This month brought a massive wave of AI releases, spanning powerful new language models, countless ComfyUI plugins, and innovative tools for audio and video generation. If you fell behind, here is your straightforward recap of everything you need to know.
DeepSeek-V4-Flash and DeepSeek-V4-Pro both arrived to handle up to one million tokens using mixture-of-experts designs for high efficiency. Qwen3.6-27B launched as an open-weight model for coding, reasoning, and multimodal tasks. XiaomiMiMo MiMo-V2.5 and MiMo-V2.5-Pro process text, images, video, and audio, managing complex reasoning with context windows up to one million tokens.
Nvidia Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 is an open multimodal AI system that processes video, audio, images, and text in a single workflow. Tencent Hy3-preview is a new large language model built for complex reasoning and coding. Kimi K2.6 is an open-source multimodal model made for extended autonomous tasks and programming workflows.
Mistral-Small-4 handles standard instructions, reasoning, and coding while processing large document windows. MiniMax-M2.7 manages complex tasks through self-directed learning and team coordination. Google Gemma 4 31B instruction-tuned processes text, images, and video while supporting extended conversations.
Google DeepMind gemma-4-26B-A4B-it and gemma-4-E4B-it both run efficiently on desktop hardware while handling text, images, and audio. LiquidAI LFM2.5-VL-450M is a compact vision-language model for fast local processing. LG AI Research EXAONE-4.5-33B combines visual understanding with strong reasoning skills.
LongCat-Next processes text, images, and audio within a single system by treating them as language tokens. Qwen3.6-35B-A3B processes text, images, and videos for automated software development. Holo3-35B-A3B lets AI programs read screens and control software across websites.
Several creators released models with removed safety filters this month. HauhauCS Qwen3.6-27B-Uncensored-HauhauCS-Aggressive, Qwen3.5-9B-Uncensored-HauhauCS-Aggressive, and the HauhauCS Qwen3.6 uncensored variant all strip standard restrictions to allow unrestricted prompts. Gemma-4-E4B-Uncensored-HauhauCS-Aggressive removes content filters from Google's four-billion parameter architecture.
OBLITERATUS gemma-4-E4B-it-OBLITERATED eliminates hard refusals from Google's Gemma 4 model. Supergemma4-26b-uncensored-gguf-v2 delivers open conversation without restrictive filters in a compressed format. Dealignai Gemma-4-31B-JANG_4M-CRACK restores unrestricted response generation while keeping core reasoning intact.
Sarvam-30b-uncensored operates without safety filters to output direct responses. Gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking removes filters and activates a built-in reasoning workflow. Nemotron3-Nano-4B-Uncensored-HauhauCS-Aggressive removes refusals from NVIDIA's Nemotron-3 Nano model.
Chaperone-Thinking-LQ-1.0 is a compressed reasoning system that scores well on medical questions while using very little memory. Talkie is a thirteen-billion parameter model trained exclusively on text published before 1931. Laguna-XS.2 is built for automated programming tasks and activates only three billion units of capacity.
inclusionAI Ling-2.6-flash runs automated workflows faster and more cheaply by keeping responses short. DaVinci-LLM delivers strong reasoning and coding skills using just three billion parameters. Darwin-4B-David and Darwin-35B-A3B-Opus both handle complex reasoning while managing multiple formats.
GigaChat 3.1 Lightning is a compact model built for fast local inference that activates only 1.8 billion parameters. IBM Granite-4.0-3B-Vision extracts structured data from documents and charts. OpenSenseNova SenseNova-U1 processes text and images through a unified architecture.
Microsoft harrier-oss-v1 converts text into dense representations to help computers understand meaning across languages. SycoFact identifies biased agreement and unsafe responses in large language outputs. Tanaos-text-summarization-v1 shortens long documents into clear sentences without losing core details.
Arcee AI Trinity-Large-Thinking generates visible reasoning steps before delivering final answers. Qwopus3.6-27B-v1-preview-GGUF prioritizes consistent formatting for local reasoning tasks. Jackrong Qwopus3.5-27B-v3-GGUF replaces lengthy pre-planning with faster execution and correction.
Qwopus-GLM-18B-Merged-GGUF combines two nine-billion parameter models into an eighteen-billion system. Lordx64 Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled replicates structured problem-solving on local machines. Jackrong Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled adds reliable step-by-step reasoning for coding assistants.
Acervo-extractor-qwen3.5-9b-GGUF pulls ordered information from invoices and legal contracts. LiquidAI LFM2.5-350M is a compact text-only model for data extraction on various hardware platforms. Zai-org GLM-5.1 focuses on maintaining accuracy during lengthy technical coding sessions.
Nemotron-3-Super-64B-A12B-Math-REAP-GGUF brings a mathematical reasoning model to local machines efficiently. Qwen3.6-27B-3bit-mlx is optimized for Apple processors, shrinking file sizes while keeping reasoning abilities. Hesamation compressed models run advanced reasoning tasks on personal computers using Qwen3.6 checkpoints.
Unsloth Qwen3.6-27B-GGUF is a locally runnable version optimized for offline coding. ByteShape Qwen3.5-9B GGUF lets developers run the system locally with low memory usage. Bonsai-8B-gguf compresses a language model into a single 1.15 GB file using one-bit packing.
Carnice-9b-W8A16-AWQ delivers an 8-bit quantized model for faster local generation. OLMo-3 7B Instruct 1-bit reduces file size to roughly one gigabyte for extreme local compression. LilaRest gemma-4-31B-it-NVFP4-turbo cuts memory usage by seventy percent while keeping original performance intact.
APEX is a new compression method that reduces file size while preserving accuracy by adjusting precision levels. TurboQuant compresses high-dimensional vectors into compact representations without needing calibration data. Llamafile delivers large language models through a single portable file that runs locally without installation.
Hipfire is an inference engine built specifically for AMD graphics hardware that processes models without traditional frameworks. DFlash and Qwen3.6-35B-A3B-DFlash speed up text generation by predicting multiple words at once. Dflash-mlx brings exact speculative decoding to Apple Silicon chips.
DMax processes multiple predictions simultaneously for faster text and code generation. Qwen3.6-27B-FP8 reduces memory usage while maintaining performance. Marco-Mini operates efficiently on consumer hardware by activating less than one billion parameters.
Z-Anime is a fully trained anime image generation model that uses everyday descriptive sentences. Flux2-Klein-9B-Consistency maintains steady visual qualities across multiple prompt runs. Nucleus-Image creates pictures using a specialized architecture that activates a small fraction of its total capacity.
Baidu ERNIE-Image generates high-quality pictures while giving users precise control over layout and text placement. LLaDA2.0-Uni brings image generation, visual analysis, and editing together into a single system. Z-Image-SAM-ControlNet transforms segmented images into photorealistic pictures.
SmartPhotoCrafter edits photographs without manual prompts by spotting visual flaws and applying corrections. RvR fixes generated images by completely redrawing them instead of attempting minor edits. SpatialEdit performs precise geometry-driven changes like moving objects and rotating items.
Patch-forcing changes image generation by applying different noise removal speeds to separate sections. Generative Refinement Networks GRN replaces standard diffusion techniques with progressive visual refinement. UDM-GRPO combines uniform discrete diffusion with reinforcement learning for text-to-image generation.
ParetoSlider lets a single image generation system handle multiple competing goals without separate training. StyleID generates identity markers resistant to artistic styling. UniGenDet combines image creation and synthetic media verification into one pipeline.
Meta-CoT improves local image editing through a two-stage thought breakdown. UniGeo edits images with precise camera movements while keeping scene boundaries intact. PixelSmile modifies specific facial expressions with precise control over intensity levels.
LumiPic transforms standard photos into high dynamic range files preserving bright and dark details. GyroScope automatically detects and corrects improperly rotated pictures. Toon-Tacular-Qwen-LoRA brings the aesthetic of late 1990s and early 2000s cartoons to AI generation.
FlowInOne turns text prompts, layouts, and editing instructions directly into visual data. Qwen3.5-4B-Base-ZitGen-V1 converts images into detailed text instructions for AI generators. See-through converts flat anime drawings into editable, multi-layered files suitable for animation.
LTX-Desktop version 1.0.5 is a stability-focused upgrade for the open-source video generation app. DisCa is an acceleration framework that speeds up AI video generation by 11.8 times. DynamicRad accelerates long video generation by applying smart sparse attention.
Motif-Video-2B transforms text prompts and static images into short video clips. Matrix-Game-3.0 generates real-time video at 720p resolution and 40 frames per second. OmniVTG-7B pinpoints exact video segments using simple text prompts.
TS-Attn improves how AI models handle videos with multiple sequential actions. LTX-2.3-22b-IC-LoRA-Outpaint expands existing footage by filling blank areas with matching material. ControlFoley transforms video clips into synchronized soundtracks.
Netflix Void-model removes video subjects while reconstructing the physical interactions they caused. Apple Ml-videoflextok converts footage into flexible sequences instead of fixed grids. LumosX generates personalized videos with multiple consistent subjects.
NVIDIA Lyra-2.0 builds walkable three-dimensional scenes from just one picture. Tencent HY-World-2.0 transforms text and media into navigable three-dimensional spaces. AnyRecon turns scattered photographs into complete three-dimensional scenes using video-based AI.
trellis-mac enables native image-to-3D model generation on Apple Silicon computers. Tencent HY-Embodied-0.5 improves spatial awareness and planning for physical robots. Lingbot-map converts continuous video feeds into accurate three-dimensional maps in real time.
WhatDreamsCost-ComfyUI simplifies video and audio editing by automating timing calculations. VibeComfy connects automated coding assistants with ComfyUI to build media pipelines locally. Comfyui-dgx-spark stabilizes ComfyUI on NVIDIA DGX Spark hardware by adjusting memory handling.
ComfyUI-Subworkflow transforms entire interface projects into reusable components. ComfyUI-ConnectTheDots simplifies complex node wiring with a dedicated control sidebar. Comfyui-Command-Palette places a fast-access command palette directly inside the workspace.
Comfy-Canvas v1.0 places a complete painting and editing workspace directly inside ComfyUI. ComfyUI-Prompt-Manager organizes, generates, and extracts text prompts for projects. ComfyUI-Majoor-AssetsManager tracks, catalogs, and previews every image or video produced locally.
SparknightLLC ComfyUI-ComboFilter streamlines cluttered dropdown menus using customized allowlists. SparknightLLC ComfyUI-GraphConstantFolder cuts prompt validation delays down to a fraction of a second. ComfyUI-Fast-Group-Bypasser-Linked introduces automated group synchronization for toggling sections.
Comfymodeldownloader automates downloading and organizing AI models by reading workflow files. Overtli Studio Suite routes local and cloud AI tasks through a single interface. ComfyUI-3D-Viewer-Pro provides a professional-grade 3D model viewer and rendering engine.
ComfyUI-Enhancement-Utils brings essential utility features with full support for nested subgraphs. Winnougan-nodes bundles essential utilities and specialized loaders. Deno2026 utility pack prioritizes everyday workflow efficiency with image resizing and batch loading.
Adonis_flux2klein cleans and upscales photographs without changing their original layout. ComfyUI-HiresFix-Ultra-AllInOne merges image upscaling, sampling, and color correction into one interface. ComfyUI_Steudio calculates ideal scaling dimensions for high-resolution enhancement.
ComfyUI-DiffAid-Patches adjusts how diffusion models process text commands during image creation. ComfyUI-NAG-Extended improves how local models process instructions by restoring reliable negative guidance. ComfyUI-Egregora-Adaptive-Colorfix aligns color tones between reference and target images.
ComfyUI-zveroboy-photo mimics the technical traits of real camera photographs. ComfyUI-Darkroom provides professional color grading and film emulation capabilities. ComfyUI-ImageViewer adds a dedicated panel to centralize image review and node inspection.
ComfyUI-Photopea-tab brings a web-based image editor directly into the workspace. Compose-Plugin-Comfyui lets users arrange multiple images on a canvas for better composition. Muffins-Flat-2-Panoramic-node transforms standard images into immersive panoramic content.
ComfyUI-Panorama-Stickers brings native video support for placing and adjusting image elements. ComfyUI-DreamScene360 converts a single panoramic image into a three-dimensional point cloud. ComfyUI_HYWorld2 transforms single photographs into three-dimensional scenes.
Comfyui-multi-seed-sampler replaces traditional random noise seeding with a fixed coordinate system. comfyui-batch-blend processes two separate video sequences and merges them together frame by frame. ComfyUI-Image-Conveyor introduces a drag-and-drop queue system for local workflows.
ComfyUI-External-Lora-Loader reads saved style models directly from external hard drives. ComfyUI-KleinRefGrid stitches up to four pictures into one unified reference grid. ComfyUI-Load-Image-Media-Browser attaches a visual file navigator directly to loading nodes.
ComfyUI_Z-Image_turbo_OPENVINO enables image generation directly through Intel integrated graphics. ComfyUI-Anima-LLLite offers a lightweight method for steering generative models using reference pictures. ComfyUI-rogala replaces manual adjustments with an embedded gallery of aesthetic presets.
IAMCCS-nodes fixes LoRA loading issues and simplifies complex video generation pipelines. ComfyUI-SmartSave-Paraquoxel reorganizes how local workflows store generated images. ComfyUI-Qwen3.5-Uncensored brings full compatibility for the modified Qwen3.5 series.
Qwen-3.5-Abliterated-Comfyui-nvfp4 enables multimodal tasks like image analysis directly in ComfyUI. ComfyUI-Wan-VACE-Prep simplifies video editing tasks for the Wan VACE model. ComfyUI-Wan-VACE-Video-Joiner automatically stitches multiple video clips together with smooth transitions.
ComfyUI-Spectrum-WAN-Proper speeds up WAN video generation by forecasting denoiser features. ComfyUI-FBnodes streamlines video workflows and adds utility functions for encoding. ComfyUI-YOLOE26 segments objects in images using simple text prompts.
ComfyUI-MurMur applies quick color styling to individual nodes using a floating palette. ComfyUI_NodeInvaders transforms the workspace into a fully playable arcade game. ComfyUI-skill-public connects an OpenClaw agent for natural language control.
lovis93 1980s monitor adapter applies authentic late-1980s monitor aesthetics directly to AI video outputs. LiconStudio Ltx2.3-VBVR-lora-I2V helps the LTX2.3 video model follow detailed motion instructions. Adetailer-hires-sync automatically enables face correction during high-resolution upsampling.
MOSS-TTS-Nano-100M is a lightweight text-to-speech engine that generates natural audio directly on standard computers. ACE-Step 1.5 XL produces complete music tracks in just eight steps. Acestep.cpp is a local AI music generation server that turns text descriptions into stereo 48kHz audio songs.
OmniVoice converts written words into spoken audio across more than six hundred languages. VoxCPM2 generates studio-quality audio from written text at higher frequencies. LongCat-AudioDiT generates high-fidelity audio directly from text inputs on the waveform latent space.
Vernacula converts audio recordings into accurate transcripts without sending data to the cloud. Trelis speech transcription handles overlapping conversations between two participants locally. Foundation-1 generates tempo-synced, key-aware loops for structured music production.
Moss Audio GFF converts audio and video files into structured text descriptions and captions. ComfyuAudioNodes-BitsAndBobs delivers new components for local audio generation and manipulation. Qwen3-TTS Easy Finetuning simplifies the process of training custom voice models.
Meeseeks Hive operates as an autonomous agent system that writes, tests, and refines code without manual oversight. Compaas transforms single users into virtual execution teams to plan, build, and deliver products. AgentOffice is an open-source workspace where AI assistants and users edit documents together.
OpenLeash acts as a safety check for autonomous AI agents before completing sensitive tasks. ToolGuard serves as a dedicated security firewall for AI agents to stop crashes before they happen. vmDeshpande's AI Agent Automation introduces dynamic decision-making for local workflow execution.
Bitterbot delivers a local-first AI agent that remembers user preferences and operates continuously. Mesh creates a shared network for running AI models across multiple computers on a local connection. TraceMind continuously monitors AI application performance and delivers real-time alerts.
Spring AI Playground runs AI agent tools in a secure, self-contained local desktop environment. Finalrun-agent tests Android and iOS applications using natural language written in YAML files. AgentHandover converts repetitive macOS routines into structured instruction files for AI agents.
SlopLobster is a self-contained local AI coding agent that operates directly on your hardware. Kon is a lightweight terminal coding assistant for everyday programming tasks. Lerim-cli operates as a background memory agent that captures important coding decisions locally.
Omni-cli is a terminal-based AI assistant that manages complex coding tasks without filling up memory. Omnix manages text, vision, speech, and audio generation entirely on personal hardware. ToolLoop is a multi-LLM agent framework that provides coding capabilities for various AI models.
Logicstamp-context transforms TypeScript projects into simple summary files for coding assistants. Corbell creates a detailed knowledge graph for projects spanning multiple code repositories. Samuraizer is a local-first knowledge engine that organizes technical documents into a searchable database.
Agensic adds a tracking and control layer to command-line workflows for developers. Yolo-gen automates the training of a combined object detection and vision-language system. CoPaw-Flash-9B-DataAnalyst-LoRA transforms compact language models into self-directed data exploration tools.
AgentScope CoPaw-Flash-9B is a focused language model built to handle automated software tasks. CWT-V5.6 replaces the standard residual stream with a structured hub-and-spoke workspace for lighter hardware. Kimi-K2.6-GGUF allows users to run a massive reasoning model entirely on local hardware.
ZPix generates and edits images using only your computer graphics card. Locally Uncensored runs text, image, and video creation on personal hardware without content restrictions. PokeClaw turns smartphones into locally operated AI assistants.
MangoLion Stretchystudio is a free 2D animation app that converts character layers into movable mesh figures. Image-MetaHub organizes, searches, and browses AI-generated media without cloud services. Smart-Comfyui-Gallery is a standalone digital asset manager designed to organize generations.
SilkStack-Image-Browser sorts, searches, and displays AI-generated artwork without uploading files. PixlStash is a self-hosted image management platform that sorts local photo libraries. Sift allows Windows users to quickly sort large libraries of media files into specific folders.
Unsloth Studio is a web interface that lets users run and train AI models locally. Modl consolidates local image generation and custom model training into a single command-line tool. Spark-dashboard provides real-time monitoring for Linux computers equipped with NVIDIA graphics cards.
Shield-82M automatically detects and removes private data from written text. OpenAI Privacy-filter scans text and removes identifying details like names and phone numbers. Model-Database-Protocol MDBP stops large language models from writing raw SQL commands.
Bordair-multimodal is an open-source test suite featuring over half a million labeled prompts to evaluate defenses against prompt injection attacks. Evalmonkey measures how well AI agents handle real-world failures in a strictly local environment. OpenEyes gives edge devices real-time environmental awareness without relying on cloud servers.
ThetaCursed Illustrious NoobAI Style Explorer lets creators preview over sixteen thousand Danbooru artist tags before generating images. Danbooru-Dataset-Filter provides a fast graphical interface for sorting large image collections. BCE-Prettybird-Nano-Math-v0.1 is a structured collection of 500 math problems built to train models on numerical reasoning.
Tstars-VTON is an open evaluation dataset designed to test virtual try-on models under realistic shopping conditions. TRIBE v2 translates videos, audio clips, and text into detailed predictions of human brain activity. Breast-cancer-detector analyzes breast ultrasound scans to identify normal tissue, benign growths, and malignant tumors.
Ai-engineering-from-scratch is a comprehensive open-source curriculum with over 260 modules teaching AI development. World Model Bench tests whether AI world models can actually think about a scene. ENMP-LoRAMerging scans multiple adaptation files and identifies components that hurt overall accuracy when merging.
MegaStyle translates text descriptions into images that share matching artistic qualities. Meta sapiens2 tracks human figures in standard photos by identifying precise joint positions. Anima-Standalone-Trainer delivers a localized, web-based interface for fine-tuning lightweight adapters.
TurboOCR converts scanned pages and screenshots into digital text at high speed. Tidbit converts articles, research papers, and images into structured text files. Scrapedown turns raw website markup into clean text files while attaching location markers.
Quizzer transforms static PDF documents into interactive, terminal-based study courses. Abook automates the book writing process through coordinated artificial intelligence agents. Simple-captioner Version 1.0.2.1 supports batch processing of images and videos to generate descriptive text.
HybridScorer uses local GPU processing to quickly organize, score, and filter massive image collections. Gen-Searcher searches the web and gathers visual references before making new images. AI Metadata Inspector delivers instant prompt extraction for local AI workflows directly through Windows file manager.
Llama-monitor provides a web-based control panel for running and tracking local large language model servers. Local-MCP-server connects offline language models directly to live web data. Webmcp connects local language models directly to online data without paid cloud services.
MothBench tracks response speed and accuracy across one hundred twenty specialized tests for local setups. Lldev.guide is a community-driven database that tracks real-world performance for local LLM inference devices. Slack app for Hugging Face sends real-time notifications about model milestones directly to Slack channels.
Flux.2-4B-Decoder-Comparator runs two versions of the FLUX.2-klein-4B image model to highlight visual differences. KupkaProd-Cinema-Pipeline transforms written screenplays into complete videos using locally hosted AI. TagForge streamlines the storage and analysis of image-text datasets for local machine learning.
Open-toys lets users build interactive talking devices locally on Apple computers. AgentScope CoPaw-Flash-9B is a focused language model built to handle automated software tasks without constant prompts. LongCat-Next is a native multimodal model capable of processing text, images, and audio within a single system.
WhatDreamsCost-ComfyUI is a free collection of custom interface nodes and ready-made workflows built for ComfyUI. The repository simplifies video and audio editing by automating timing calculations inside a visual workspace. […]
VibeComfy connects automated coding assistants with ComfyUI to build and run media pipelines locally. The tool converts standard generation graphs into a single editable format that processes text commands instead […]
The Comfyui-dgx-spark project provides scripts and source patches that stabilize ComfyUI on NVIDIA DGX Spark hardware. By adjusting memory handling and hardware parameters, the collection prevents system freezes during image […]
Adonis_flux2klein is a specialized enhancement tool that cleans and upscales photographs without changing their original layout. It focuses on sharpening skin, hair, and structural details while keeping the subject’s exact […]
LTX-Desktop version 1.0.5 arrives as a stability-focused upgrade for the open-source video generation app, prioritizing performance tuning and interface repairs. The release delivers targeted fixes for memory management, timeline editing, […]
ThetaCursed released the Illustrious NoobAI Style Explorer, a visual library that lets creators preview over sixteen thousand Danbooru artist tags before generating images. The interface shows style references, dataset strength […]
Moss Audio GFF is a desktop application that converts audio and video files into structured text descriptions and captions. The software processes sound inputs ranging from podcasts and meetings to […]
inclusionAI just released Ling-2.6-flash, an artificial intelligence model built to run automated workflows faster and more cheaply. It handles multi-step tasks by keeping responses short while maintaining strong accuracy. Traditional […]
Laguna-XS.2 is an open source language model built specifically for automated programming tasks and extended project workflows. Built with a modular design, the system activates only three billion units of […]
Nvidia recently released Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16, an open multimodal AI system that processes video, audio, images, and text in a single workflow. Users can run it locally to summarize lengthy meetings, transcribe […]
Talkie introduces a collection of thirteen-billion parameter language models trained exclusively on text published before 1931. The accompanying Python library handles weight downloads and generates text through a straightforward command […]
MiMo-V2.5-Pro is an open-source language model built with a Mixture-of-Experts (MoE) design, handling up to one million tokens of input at once. It manages complex software workflows while keeping memory […]
XiaomiMiMo launched MiMo-V2.5, a system that processes text, images, video, and audio through one unified model. It manages complex reasoning while accepting context windows up to one million tokens. The […]
Chaperone-Thinking-LQ-1.0 is a compressed reasoning system that scores 84 percent on medical question sets while shrinking its memory footprint to roughly twenty gigabytes. The release applies targeted compression and domain […]
Meeseeks Hive operates as an autonomous agent system that writes, tests, and refines code without manual oversight. It runs continuous loops to evaluate outputs against set quality standards. Abraham Casanova […]
The Nemotron-3-Super-64B-A12B-Math-REAP-GGUF brings a compressed version of a mathematical reasoning model directly to local machines. This specialized file format allows the system to run efficiently on standard consumer hardware without […]
Shield-82M is a lightweight tool that automatically detects and removes private data from written text. It identifies over seventy sensitive categories across multiple languages, including phone numbers, financial details, and […]
Hipfire is an inference engine built specifically for AMD graphics hardware. It processes language models without relying on traditional frameworks, delivering a streamlined runtime. The tool uses simple terminal commands […]
Qwen3.6-27B-3bit-mlx offers a streamlined language model optimized specifically for Apple processors. Shrinking file sizes while keeping reasoning abilities, it runs text generation tasks smoothly on standard laptops. Creator leonsarmiento designed […]
TurboOCR operates as a dedicated recognition server that converts scanned pages and screenshots into digital text at high speed. The software handles printed material and handwritten notes using specialized graphics […]
RvR introduces a new approach to fixing generated images by completely redrawing them instead of attempting minor edits. Researchers at Tsinghua University and Tencent Hunyuan developed this system to solve […]
OmniVTG-7B is an open-source model that pinpoints exact video segments using simple text prompts. Rather than tagging entire clips, it scans long footage and marks precise start and end times […]
Z-Anime delivers a fully trained anime image generation model that operates independently from lightweight add-on patches. The system produces detailed illustrations using everyday descriptive sentences instead of strict keyword lists. […]
Danbooru-Dataset-Filter provides a fast graphical interface for sorting and organizing large image collections used in machine learning projects. The software processes millions of files in seconds, allowing users to build […]
Apple researchers released Ml-videoflextok, an open source video tool that converts footage into flexible sequences instead of fixed grids. This approach stores broad motion first, then adds sharper visual details […]
Generative Refinement Networks, or GRN, offers a new method for creating digital images and video. The approach replaces standard diffusion techniques with progressive visual refinement and adaptive computing. Built by […]
DisCa is a new acceleration framework designed to speed up AI video generation models. It reduces processing time by 11.8 times without sacrificing visual clarity. Tencent researchers who also made […]
ControlFoley transforms video clips into synchronized soundtracks by combining visual scenes, written descriptions, and existing audio samples into a single generation system. This new framework produces matching sound effects and […]
Combining specialized AI adapters into one system now works best when harmful layers are removed first. The ENMP-LoRAMerging project scans multiple adaptation files, identifies components that hurt overall accuracy, and […]
Yovecent has released UDM-GRPO, an open-source framework that combines uniform discrete diffusion with reinforcement learning for text-to-image generation. The system stabilizes training and improves output quality by treating the fully […]
Tencent researchers recently published MegaStyle, a system designed to automate the creation of visual style libraries. The pipeline translates text descriptions into images that share matching artistic qualities while keeping […]
Tstars-VTON is an open evaluation dataset designed to test virtual try-on models under realistic shopping conditions. It contains 1,780 image pairs covering layered clothing, footwear, and accessories across dozens of […]
SmartPhotoCrafter is an open-source framework that edits photographs without requiring manual prompts. The system automatically spots visual flaws, plans specific improvements, and applies corrections in a single continuous workflow. Researchers […]
AnyRecon turns scattered photographs into complete three-dimensional scenes using a video-based artificial intelligence system. The framework processes inputs in any order without needing precise spacing between camera angles. OpenImagingLab built […]
TS-Attn introduces a new attention method that improves how AI models handle videos with multiple sequential actions. The system rearranges how the model focuses on time-based data, allowing complex prompts […]
Patch-forcing changes how artificial intelligence generates images by applying different noise removal speeds to separate sections of a picture. Easier areas process quickly while complex sections receive additional refinement, making […]
DynamicRad accelerates long video generation by applying smart sparse attention to existing AI diffusion models. This open framework cuts processing time substantially while keeping visual quality consistent across full-length clips. […]
Meta researchers have released sapiens2, a vision model collection designed to track human figures in standard photos. The software identifies precise joint positions, maps anatomical sections, and estimates surface angles […]
ParetoSlider lets a single image generation system handle multiple competing goals without needing separate training runs. By adjusting a simple preference setting during creation, users can smoothly shift between visual […]
UniGenDet combines artificial intelligence image creation and synthetic media verification into a single operational pipeline. The framework processes both workflows simultaneously to share quality signals in real time. Developed by […]
StyleID is a specialized image encoder built on the CLIP framework that generates identity markers resistant to artistic styling. Instead of struggling when faces become cartoons, the tool maintains consistent […]
Anima-Standalone-Trainer delivers a localized, web-based interface for fine-tuning lightweight adapters on the Anima image generation system. The project operates independently of larger ecosystems, keeping the workflow separate for local deployments. […]
OpenSenseNova released SenseNova-U1, an open-source model family designed to process text and images through a single unified architecture. Unlike older systems that patch together separate vision and language parts, this […]
LumiPic transforms standard digital photos into high dynamic range files that preserve visible details in both extreme bright spots and deep shadows. The tool works as a lightweight add-on that […]
UniGeo offers a structured approach to editing images with precise camera movements while keeping original scene boundaries intact. The framework adjusts perspective angles and focal paths without distorting the underlying […]
Meta-CoT introduces a structured reasoning framework designed to improve local image editing workflows. The open-source model processes user instructions through a two-stage thought breakdown, separating editing goals into specific actions […]
SparknightLLC has released ComfyUI-ComboFilter, a frontend extension that streamlines cluttered dropdown menus inside the interface. By applying customized allowlists and blocklists to widget selections, the tool automatically removes irrelevant options […]
ComfyUI-Load-Image-Media-Browser attaches a visual file navigator directly to ComfyUI’s standard image and video loading nodes. Users can preview, sort, and pick media files without leaving the main workspace or manually […]
ComfyUI-Subworkflow introduces a modular system that transforms entire interface projects into reusable components. This extension runs complex pipelines from a single node while managing clear input boundaries. Developer Eniewold released […]
ComfyUI_Z-Image_turbo_OPENVINO is a custom node that enables image generation directly through Intel integrated graphics. Routing inference tasks through the OpenVINO framework reduces standard processing times to roughly ninety seconds per […]
ComfyUI-Anima-LLLite introduces a lightweight method for steering generative image models using reference pictures. The custom node attaches trained adjustments to the Anima architecture, allowing users to guide how the software […]
ComfyUI_NodeInvaders transforms a standard image generation workspace into a fully playable arcade experience. This custom extension replaces your usual creative canvas with an interactive shooter where you target enemy sprites […]
Comfyui-Command-Palette, a new interface extension for ComfyUI places a fast-access command palette directly inside the workspace, allowing users to run actions and navigate graphs without relying on standard menus. This […]
ComfyUI-rogala introduces a custom node pack that streamlines prompt creation and video sampling tasks. The toolkit replaces manual adjustments with an embedded gallery that applies aesthetic presets directly to model […]
A new tool called Comfymodeldownloader automates downloading and organizing AI models for ComfyUI workflows. Dropping a single workflow file into the interface triggers automatic placement, removing the need for manual […]
The Qwopus3.6-27B-v1-preview-GGUF model provides a locally usable checkpoint built directly on the Qwen3.6-27B reasoning architecture. The release prioritizes consistent formatting and reduces stylistic drift during long conversations. Jackrong applied a […]
Lordx64 has released Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled that replicates structured problem-solving while operating entirely on local machines. The system uses expert routing to produce detailed thinking steps before delivering answers, which makes technical […]
The recently released Kimi-K2.6-GGUF format allows users to run a massive open-source reasoning model entirely on local hardware. This version supports long-form programming, image analysis, and video processing while keeping […]
ComfyUI-NAG-Extended is a recent software extension that improves how local image and video generation models process instructions. By restoring reliable negative guidance, the tool gives creators more precise control over […]
Qwen3.6-35B-A3B-DFlash acts as a support component designed to dramatically accelerate text generation for large language models. It works by drafting several words at once before the main system finishes processing […]
Qwen3.6-27B-FP8 delivers an open-weight artificial intelligence model compressed using an efficient FP8 format to reduce memory usage while maintaining performance. It processes text, images, and video simultaneously to handle complex […]
IAMCCS-nodes introduces and updates a set of custom modules designed to fix LoRA loading issues and simplify complex video generation pipelines inside ComfyUI. The toolkit automatically remaps model weights and […]
Motif-Video-2B is an open-source model that transforms text prompts and static images into short video clips. Built with just two billion parameters, it delivers competitive generation results while using significantly […]
Tencent recently released Hy3-preview, a large language model designed to handle complex reasoning, coding, and automated tasks with improved accuracy. The system uses a mixture-of-experts layout that activates only a […]
Hesamation recently released compressed model files designed to run advanced reasoning tasks on personal computers. These files translate a Qwen3.6 checkpoint trained to follow the step-by-step problem-solving style of Claude […]
HauhauCS recently released the Qwen3.6-27B-Uncensored-HauhauCS-Aggressive model, a refined version of the original architecture that removes all standard safety filters. This iteration delivers complete compliance with complex instructions, achieving zero recorded […]
LLaDA2.0-Uni brings image generation, visual analysis, and editing together into a single downloadable system. The framework processes text and visual data through a unified diffusion design, removing the need to […]
The Unsloth team recently quantized Qwen3.6-27B-GGUF, a locally runnable version of a language model optimized for offline coding tasks. Like Qwen3-Coder-Next also quantized by the Unsloth team, this quantized package […]
DeepSeek-AI has released a preview version of its new language model, DeepSeek-V4-Flash, which processes up to one million tokens while maintaining high efficiency. The system operates as a mixture-of-experts network, […]
Qwen3.6-27B is an open-weight artificial intelligence model designed for coding, reasoning, and multimodal tasks. The system processes large documents and media at once while handling extensive conversational history. Released by […]
OpenAI recently launched Privacy-filter, a machine learning tool designed to automatically scan text and remove identifying details like names, emails, and phone numbers. The system processes information in a single […]
DeepSeek-V4-Pro is a new open-source language model capable of processing up to one million tokens in a single prompt. The architecture uses a mixture-of-experts (MoE) design, which activates only a […]
ComfyUI-Majoor-AssetsManager serves as an integrated file browser that tracks, catalogs, and previews every image or video produced within your local generation environment. The extension automatically indexes new outputs while providing […]
ComfyUI-ConnectTheDots is a straightforward visual workflow extension for the ComfyUI platform that simplifies complex node wiring. It replaces tedious manual canvas panning with a dedicated control sidebar that locates matching […]
Muffins-Flat-2-Panoramic-node provides a set of custom tools for ComfyUI that transform standard images and video frames into immersive panoramic content. The package automates an outpainting workflow that places flat visuals […]
Evalmonkey provides a strictly local testing environment that measures how well artificial intelligence agents handle real-world failures. Instead of relying on perfect conditions, the framework deliberately breaks normal operations to […]
Yolo-gen automates the training of a combined object detection and vision-language system using a single command. The software reads a standard detection dataset, trains the primary model, and automatically creates […]
Local-MCP-server connects offline language models directly to live web data. The lightweight program acts as a bridge that allows existing software to run online searches, save webpage snapshots, and pull […]
Spark-dashboard provides real-time monitoring tools for Linux computers equipped with NVIDIA graphics cards. The software tracks hardware performance and local model inference metrics from a single browser interface. Independent developer […]
ComfyUI-External-Lora-Loader removes the restriction that forces users to keep style files in a single installation folder. This custom node reads saved style models directly from external hard drives, network storage, […]
ComfyUI-KleinRefGrid replaces multiple reference image nodes with a single custom component that stitches up to four pictures into one unified grid. xb1n0ry created this utility to streamline the image conditioning […]
A new open-source adapter now applies authentic late-1980s monitor aesthetics directly to AI video outputs. Created by developer lovis93, this file modifies the standard video model to simulate scanlines, signal […]
Modl consolidates local image generation and custom model training into a single command-line tool that removes complex setup steps. The program handles everything from downloading large files to running inference […]
Deno2026 recently released a utility pack for ComfyUI that prioritizes everyday workflow efficiency over experimental features. The package delivers three focused custom nodes designed to handle image resizing, batch loading, […]
ComfyUI-Panorama-Stickers recently received a major update that brings native video support directly into local image editing workflows. This open-source extension allows users to place, adjust, and composite image elements onto […]
ComfyUI-DiffAid-Patches introduces two custom nodes that adjust how diffusion models process text commands during image creation. The tool modifies guidance strength across specific network sections instead of relying on a […]
Kimi K2.6 launches as an open-source multimodal model built for extended autonomous tasks and complex programming workflows. It processes lengthy instructions and coordinates multiple sub-tasks to deliver complete outputs from […]
Lerim-cli operates as a background memory agent designed specifically for automated coding workflows. It runs alongside your development tools to capture important decisions and continuously store them locally as plain […]
Omnix operates as a local AI environment that manages text, vision, speech, and audio generation entirely on personal hardware. The system routes tasks through a lightweight controller to automatically load […]
Omni-cli is a terminal-based AI assistant designed to manage complex coding tasks without filling up system memory. It connects to various language models directly from the command line, allowing users […]
OpenLeash is an open-source authorization system that acts as a safety check for autonomous AI agents. Before an AI tool completes a sensitive task like making a purchase or sending […]
CWT-V5.6 replaces the standard undifferentiated residual stream in typical models with a structured hub-and-spoke workspace designed to run efficiently on lighter hardware. This shift lets the system selectively manage memory […]
Mistral-Small-4 operates as a single system that handles standard instructions, complex reasoning, and coding tasks. The architecture processes large document windows while accepting both text and visual inputs. Mistral AI […]
The new trellis-mac project enables native image-to-3D model generation on Apple Silicon computers. By adapting a previously Windows-only pipeline, the tool converts single photographs into detailed mesh assets with physically […]
Trelis recently released a specialized speech transcription model that handles overlapping conversations between two participants. The system processes audio clips locally without relying on external cloud servers. Built as an […]
ComfyuAudioNodes-BitsAndBobs delivers a new collection of ComfyUI components designed for local audio generation and manipulation. The update adds native support for quantized models and introduces several pathways to inject reference […]
ComfyUI-MurMur introduces a lightweight interface add-on that applies quick color styling to individual nodes and workflow groups. Users simply press the Tab key to open a floating palette, select a […]
ComfyUI_HYWorld2 brings custom nodes to a popular image generation interface, allowing users to transform single photographs or panoramic shots into three-dimensional scenes. The extension processes visual inputs to output editable […]
Flux2-Klein-9B-Consistency is a fine-tuned image generation model designed to maintain steady visual qualities across multiple prompt runs. The latest version automatically resolves common color distortion problems while delivering cleaner results […]
SparknightLLC has released ComfyUI-GraphConstantFolder, a server-side extension that cuts prompt validation delays from roughly one second down to a fraction of that time. The tool automatically rewrites submitted workflow graphs […]
ComfyUI-Fast-Group-Bypasser-Linked introduces automated group synchronization to an existing visual workflow tool. A single JavaScript file now enables connected toggling for multiple workflow sections without altering original code or requiring compilation. […]
Comfy-Canvas v1.0 places a complete painting and editing workspace directly inside ComfyUI. The tool removes the need for extra browser tabs by running a built-in overlay for masks, brushes, and […]
ComfyUI-Prompt-Manager is a local toolkit designed to organize, generate, and extract text prompts and complete generation settings for ComfyUI projects. The software replaces scattered note-taking and manual node copying with […]
ZPix is a local application that generates and edits images using only your computer graphics card. The software processes visual tasks directly on your hardware while maintaining a clean interface. […]
Lingbot-map converts continuous video feeds into accurate three dimensional maps in real time. The system tracks camera movement and builds spatial grids as footage records. Robbyant created this package to […]
The Qwopus-GLM-18B-Merged-GGUF combines two separate nine-billion-parameter models into a single eighteen-billion-parameter system. By stacking thirty-two layers from each version, it creates a deeper network optimized for local reasoning tasks and […]
Nucleus-Image creates pictures from text prompts using a specialized architecture that activates a small fraction of its total capacity during each run. The system divides tasks across multiple routing layers […]
NVIDIA recently released Lyra-2.0, a system that builds walkable three-dimensional scenes from just one picture. The framework creates long camera videos that maintain consistent geometry before turning them into explorable […]
HauhauCS has released an uncensored variant of the Qwen3.6 (shortly after their 3.5 variant Qwen3.5-9B-Uncensored-HauhauCS-Aggressive) a language model that removes standard content restrictions. The system processes inputs through an expert-routing […]
OBLITERATUS recently released v3 of gemma-4-E4B-it-OBLITERATED, a modified version of Google’s Gemma 4 designed to eliminate hard refusals. The 4B parameter model strips away built-in content filters while preserving core […]
Tencent has released HY-World-2.0, an open-source framework that transforms text, single photographs, and video clips into navigable three-dimensional spaces. Rather than generating short playback loops, the system builds complete digital […]
Qwen3.6-35B-A3B is an open-weight language model designed for automated software development and complex task management. The system processes text, images, and videos while maintaining a continuous reasoning process before generating […]
The recently released Qwen3.5-9B-Uncensored-HauhauCS-Aggressive removes standard safety filters from the original nine-billion-parameter AI model while keeping all original text and image processing abilities intact. Users can input unrestricted prompts without […]
Jackrong recently released Qwopus3.5-27B-v3-GGUF, a local model that replaces lengthy pre-planning with faster execution and iterative correction. This structure reduces token consumption while maintaining strong performance on coding tasks. Built […]
AgentOffice is an open-source workspace that allows AI assistants and users to edit documents, databases, presentations, and diagrams in a shared environment. Instead of treating automation scripts as background helpers, […]
GyroScope is a neural network that automatically detects and corrects improperly rotated pictures. It identifies whether an image faces zero, ninety, one hundred eighty, or two hundred seventy degrees up […]
SlopLobster is a self-contained local AI coding agent that operates directly on your hardware without relying on external cloud services. The program reads, modifies, and tracks project files while executing […]
The Carnice-9b-W8A16-AWQ release delivers an 8-bit quantized version of a text-focused language model optimized for local deployment. By compressing the original architecture, it enables faster generation while using less memory […]
The OLMo-3 7B Instruct model now operates at a 1-bit precision level through a recent experimental release. This extreme compression reduces the file to roughly one gigabyte, allowing local deployment […]
Dflash-mlx brings exact speculative decoding to modern Silicon chips using Apple’s MLX framework. A smaller draft network predicts several words ahead, then confirms them instantly to speed up generation while […]
Comfyui-multi-seed-sampler introduces a new approach to generating images that replaces traditional random noise seeding with a fixed coordinate system. By treating generation as a predictable path through structured data, the […]
comfyui-batch-blend processes two separate video sequences or image sets and merges them together frame by frame without relying on complex workarounds. The custom node performs these operations directly on data […]
ComfyUI-Photopea-tab brings a web-based image editor directly into the ComfyUI workspace through a new sidebar extension. The tool embeds Photopea into the main interface, allowing creators to move images between […]
ComfyUI-Image-Conveyor introduces a drag-and-drop queue system designed specifically for local image generation workflows. This custom node lets users drop multiple pictures into a single interface element and feeds them into […]
Image-MetaHub is a desktop application designed to organize, search, and browse AI-generated media without relying on cloud services. It indexes local folders, reads embedded generation data, and provides fast filtering […]
Winnougan-nodes is a new collection of custom nodes designed to streamline image and video generation workflows within ComfyUI. The package bundles essential utilities, specialized loaders, and optimized samplers into a […]
Video creators can now expand existing footage using a new training module that fills designated blank areas with matching visual material. LTX-2.3-22b-IC-LoRA-Outpaint identifies pure black sections in an input clip […]
SpatialEdit is an open source research model that performs precise, geometry-driven changes to static images. It moves objects, rotates items, and shifts camera angles while keeping the original scene and […]
MangoLion recently released Stretchystudio, a free 2D animation app that converts character layers into movable mesh figures in seconds. The software connects AI image outputs directly to actual animation workflows. […]
Flux.2-4B-Decoder-Comparator is a local testing application that runs two versions of the FLUX.2-klein-4B image model at the same time to highlight visual differences. It places output from a standard decoder […]
Compaas is an open-source platform that transforms single users into virtual execution teams. It coordinates specialized artificial intelligence agents to plan, build, test, and deliver digital products from a single […]
The Sarvam-30b-uncensored project delivers a modified version of the original Sarvam-30B system, specifically engineered to operate without standard safety filters. By removing internal alignment constraints, the architecture outputs direct responses […]
TraceMind functions as an open-source observability platform that continuously monitors AI application performance. The system automatically tracks output quality, flags hidden regressions, and delivers real-time alerts when response standards decline. […]
Spring AI Playground is a cross-platform desktop application that runs AI agent tools in a secure, self-contained local environment. The software focuses on letting users build, test, and validate Model […]
Bitterbot delivers a local-first artificial intelligence agent that remembers user preferences and operates continuously in the background. Unlike standard chatbots that erase context after a session ends, this system stores […]
Mesh creates a shared network for running artificial intelligence models across multiple computers on a local connection. The system divides files into pieces and sends tasks directly between devices instead […]
Tidbit is a command-line utility that converts articles, research papers, ebooks, and images into structured text files and training-ready data logs. The tool processes user-provided templates to pull exact information […]
Alibaba International Digital Commerce recently released Marco-Mini, a multilingual language system built to operate efficiently on standard consumer hardware. The architecture processes text by activating only 0.86 billion parameters out […]
Overtli Studio Suite introduces a unified collection of custom nodes that route local and cloud AI tasks through a single ComfyUI interface. The tool consolidates text generation, media creation, and […]
BCE-Prettybird-Nano-Math-v0.1 is a structured collection of 500 math problems and answers built to train language models on numerical reasoning. The dataset links specific prompts with input values and expected results, […]
Webmcp connects local language models directly to online data without routing queries through paid cloud services. AuthBits released the project for users who prefer private automation on personal hardware. Running […]
TRIBE v2 is an open-source software framework that translates videos, audio clips, and written text into detailed predictions of human brain activity. The system uses machine learning to map sensory […]
Kon is a lightweight terminal coding assistant designed to keep your working context small while handling everyday programming tasks. The tool manages files, runs commands, and searches projects through a […]
Bordair-multimodal is an open-source test suite featuring over half a million labeled prompts built to evaluate defenses against prompt injection attacks. The collection proves that splitting harmful instructions across multiple […]
CoPaw-Flash-9B-DataAnalyst-LoRA transforms compact nine-billion-parameter language models into self-directed data exploration tools. The adapter handles file loading, statistical profiling, chart creation, and automated Python scripting without requiring repeated manual prompts to […]
Locally Uncensored is a desktop program that runs text, image, and video creation on personal hardware. The tool handles model downloads, detects installed backends, and operates without content restrictions. Developers […]
DMax delivers faster text and code generation by processing multiple predictions simultaneously while maintaining output quality. The system handles parallel computation steps and automatically fixes errors before they spread through […]
ComfyUI-ImageViewer adds a dedicated panel to the workspace that centralizes image review, playback, and node inspection. Users route outputs into separate tabs, scrub timelines, and examine masks outside standard preview […]
FlowInOne transforms image generation by turning text prompts, layouts, and editing instructions directly into visual data. Instead of juggling specialized tools, users route all requests through a single system that […]
Qwen3.5-4B-Base-ZitGen-V1 is a lightweight, fine-tuned model built to convert images into detailed text instructions for AI generators. It focuses on producing highly specific commands optimized for Z-Image Turbo workflows. Independent […]
Darwin-4B-David processes complex reasoning tasks in a 128,000-token window while supporting over one hundred languages. The system uses a dedicated thinking mode that breaks down difficult prompts into clear, verifiable […]
Model-Database-Protocol (MDBP) provides a secure middle layer that stops large language models from writing raw SQL commands. Instead of generating database queries directly, AI tools now send structured JSON requests […]
OpenEyes is an open-source vision framework designed to give edge devices real-time environmental awareness without relying on cloud servers. The system runs detection, tracking, depth mapping, and control tasks entirely […]
Abook is a self-hosted web application designed to automate the book writing process through coordinated artificial intelligence agents. Four specialized digital workers manage outlining, drafting, editing, and cross-referencing chapters while […]
Scrapedown is a lightweight coding package that turns raw website markup into clean text files while attaching location markers for every visible element. These tags allow language models to draft […]
Quizzer transforms static PDF documents into interactive, terminal-based study courses. Users drop a textbook, manual, or lecture notes into the tool, and it automatically generates practice questions for active recall. […]
PokeClaw is an open-source Android application that turns smartphones into locally operated AI assistants. Instead of sending data to external servers, the app processes requests directly on your device using […]
Local model testing now runs through an open-standard evaluation suite designed specifically for personal computer setups. MothBench tracks response speed, initial delay, and output accuracy across one hundred twenty specialized […]
daVinci-LLM is a new open-source language model that delivers strong reasoning and coding skills using just three billion parameters. It follows a structured two-part training approach across roughly eight trillion […]
MOSS-TTS-Nano-100M is a lightweight, open-source text-to-speech engine that generates natural audio directly on standard computers. The system converts typed prompts into clear speech while maintaining strict efficiency for daily use. […]
LiconStudio recently released Ltx2.3-VBVR-lora-I2V, a supplemental tool that help the LTX2.3 video model follow detailed instructions about motion and object placement. Instead of generating unpredictable footage, the updated weights guide […]
LiquidAI has published LFM2.5-VL-450M, a compact vision-language model built for fast local processing of images and video streams. The system processes visual inputs alongside text prompts to generate captions, detect […]
The new Gemma-4-E4B-Uncensored-HauhauCS-Aggressive removes standard content filters from Google’s four-billion parameter architecture. It generates complete responses without blocking requests while keeping every original capability intact. Independent creator HauhauCS developed this […]
LG AI Research has released EXAONE-4.5-33B, an open-weight vision language model designed to process both images and text. The system combines native visual understanding with strong reasoning skills across a […]
Google DeepMind has released the gemma-4-26B-A4B-it model, a new local AI system that processes text, images, and video while running efficiently on standard desktop hardware. This instruction-tuned version uses a […]
The gemma-4-E4B-it release brings a compact, instruction-tuned language model to the open source ecosystem. It handles text, images, and audio inputs while producing detailed written responses on standard hardware. Google […]
Jackrong released Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled that adds reliable step-by-step reasoning and fixes formatting crashes in local coding assistants. The system processes complex prompts and maintains active thinking modes for extended tasks without […]
OmniVoice is an open source text-to-speech system that converts written words into spoken audio across more than six hundred languages. The software enables instant voice matching and allows users to […]
Netflix recently released Void-model, an open framework that removes video subjects while reconstructing the physical interactions they caused. Standard editors only fill background pixels, but this tool calculates how nearby […]
LilaRest recently released gemma-4-31B-it-NVFP4-turbo, a text-only version of a large language model that cuts memory usage by nearly seventy percent while keeping original performance intact. The updated system runs smoothly […]
Baidu recently released ERNIE-Image, an open text-to-image model built on a single-stream Diffusion Transformer. The system generates high-quality pictures while giving users precise control over layout, text placement, and object […]
Supergemma4-26b-uncensored-gguf-v2 is a compressed language model that delivers open conversation without restrictive safety filters. The package wraps a 26-billion-parameter network into a GGUF container for straightforward local execution. Jiunsong developed […]
The latest release from Dealignai, Gemma-4-31B-JANG_4M-CRACK modifies the Gemma 4 31B framework to restore unrestricted response generation. This updated model removes standard safety filters while maintaining its core reasoning and […]
VoxCPM2 is an open text-to-speech system that generates studio-quality audio from written text. The tool reads words at standard clarity and outputs polished speech at a higher frequency without needing […]
Google has released the instruction-tuned version of its Gemma 4 model, offering a 31-billion parameter system that processes text, images, and video while supporting extended conversations. Users can now run […]
Tencent has released HY-Embodied-0.5, an open-source toolkit that improves spatial awareness and planning for physical robots. The system blends image recognition with logical steps so machines can safely interact with […]
MiniMax-M2.7 arrives as a new publicly released language model designed to manage complex tasks through self-directed learning and automated team coordination. The system handles coding projects, document editing, and system […]
Vernacula is a local speech processing library that converts audio recordings into accurate transcripts without sending data to the cloud. The software combines speech recognition, speaker separation, and noise removal […]
tanaos-text-summarization-v1 is a compact AI system designed to shorten long documents into clear, readable sentences without losing core details. It processes English text quickly using a rewriting approach that focuses […]
Llama-monitor provides a web-based control panel for running and tracking local large language model servers in real time. This lightweight utility tracks hardware performance, manages configuration files, and keeps system […]
DFlash introduces a fast drafting method that speeds up how large language models generate text on local machines. It uses a compact diffusion approach to predict multiple words at once, […]
AgentHandover observes daily computer activity on macOS and automatically converts repetitive routines into structured instruction files that artificial intelligence agents can follow. The system tracks mouse movements, application states, and […]
Zai-org released GLM-5.1 to handle extended coding and automation workflows. This update focuses on maintaining accuracy during lengthy technical sessions rather than losing momentum after a quick start. Researchers scaled […]
ComfyUI-SmartSave-Paraquoxel is a custom extension that reorganizes how local workflows store generated images and video sequences. The tool replaces basic export functions with interactive canvas controls that manage single files […]
ComfyUI-Qwen3.5-Uncensored brings full compatibility for the Qwen3.5 language model series to local ComfyUI workflows. The custom node scans existing folders and automatically prepares both standard and modified versions for immediate […]
ComfyUI-Egregora-Adaptive-Colorfix is a custom processing node that aligns color tones between reference and target images without distorting edges. The tool actively solves blending errors that typically appear during tiled upscaling, […]
AI Metadata Inspector delivers instant prompt extraction for local AI workflows. Users simply right-click supported image or video files to pull full generation details directly through the Windows file manager. […]
ThetaCursed has released ComfyUI-HiresFix-Ultra-AllInOne, a custom node that merges image upscaling, sampling, and color correction into one interface. This tool handles large resolutions efficiently by splitting memory loads into smaller […]
ComfyUI-zveroboy-photo is a custom node suite designed for post-processing AI-generated images to mimic the technical traits of real camera photographs. The tool focuses on sensor noise simulation, raw file handling, […]
ComfyUI_Steudio introduces a structured approach to high-resolution image enhancement. The suite calculates ideal scaling dimensions, splits pictures into sections, processes them independently, and stitches results together without borders. Built for […]
Smart-Comfyui-Gallery is a standalone digital asset manager designed to organize and search ComfyUI generations independently. The tool runs entirely outside the main application environment, allowing creators to manage thousands of […]
Local AI creators now have a dedicated offline manager designed to sort, search, and display AI-generated artwork without uploading files to external servers. SilkStack-Image-Browser parses generation metadata directly from prompts, […]
ComfyUI-Wan-VACE-Prep delivers a focused set of nodes that simplify video editing tasks for the Wan VACE model. Users can manage outpainting, clip transitions, and sequence extensions through a clean interface […]
KupkaProd-Cinema-Pipeline transforms written prompts or screenplays into complete videos using locally hosted AI models. The system handles every stage of production, including script breakdown, storyboard generation, multi-take filming, and final […]
Acestep.cpp is a local AI music generation server that converts text descriptions into complete songs. Users simply describe what they want, add lyrics if needed, and receive stereo 48kHz audio […]
ACE-Step recently published ACE-Step 1.5 XL, an open audio generation model that produces complete music tracks in just eight steps. This streamlined process significantly reduces rendering wait times while preserving […]
A newly released open weight model, Gemma-4-31B-it-Mystery-Fine-Tune-HERETIC-UNCENSORED-Thinking, removes standard content filters from a 31B parameter architecture while activating a built-in reasoning workflow. This configuration produces step-by-step internal logic before generating […]
A new three-class image classification model named Breast-cancer-detector analyzes breast ultrasound scans to identify normal tissue, benign growths, and malignant tumors. The system processes clean medical images to highlight areas […]
Logicstamp-context transforms TypeScript projects into simple summary files. Instead of feeding entire code files to automated coding assistants, this program extracts exact rules, connections, and interface layouts that tools can […]
Open-toys is a local voice intelligence application that lets users build interactive talking devices without relying on external servers. The software runs entirely on Apple computers, ensuring conversations remain private […]
Agensic adds a tracking and control layer to command-line workflows for developers. The tool monitors AI agent activity, separates human inputs from automated tasks, and delivers instant terminal suggestions. Alessio […]
Samuraizer is a local-first knowledge engine that organizes technical documents, code repositories, and research articles into a searchable database. The latest release shifts all artificial intelligence processing to your own […]
ToolGuard serves as a dedicated security firewall for AI agents, designed to stop crashes before they happen. It intercepts execution errors, such as incorrect data types or missing JSON keys, […]
Corbell is a command-line tool that creates a detailed knowledge graph for projects spanning multiple code repositories. It maps service dependencies, method signatures, and database connections directly from source code […]
ToolLoop is a multi-LLM agent framework that provides coding capabilities for a wide variety of AI models. It allows users to perform tasks like file editing, code search, and shell […]
Finalrun-agent is an AI-driven command line tool that tests Android and iOS applications using natural language. Users write tests in plain English within YAML files, and the tool launches the […]
llmdev.guide is a new community-driven database designed to track real-world performance for local LLM inference devices. It collects data to help users choose the right hardware for running large language […]
LongCat-Next is a native multimodal model capable of processing text, images, and audio within a single system. It treats visual and audio signals as language tokens, allowing the model to […]
LFM2.5-350M is a compact AI model created by Liquid AI for on-device deployment across various hardware platforms. The 350-million parameter text-only model delivers competitive performance for data extraction and structured […]
ByteShape recently published a GGUF-formatted version of the Qwen3.5-9B language model. The release enables developers to run the system locally while keeping memory usage low through optimized file compression. The […]
Bonsai-8B-gguf delivers a language model compressed into a single 1.15 GB file. It operates through standard inference engines by packing every weight into one digital bit. This approach removes typical […]
Holo3-35B-A3B is a new open model designed to let AI programs read screens and control software across websites and desktop applications. Instead of processing entire interfaces at once, it reads […]
Ai-engineering-from-scratch is a comprehensive, open-source curriculum designed to teach artificial intelligence development from the ground up. Spanning over 260 structured modules, the repository guides learners through mathematical foundations, deep learning, […]
Darwin-35B-A3B-Opus is an open language model that merges advanced reasoning with image and video understanding. The system activates just three billion parameters during operation, keeping response times quick while handling […]
The Acervo-extractor-qwen3.5-9b-GGUF is a compressed version of a nine-billion parameter model built to pull ordered information from invoices, legal contracts, and financial reports. By converting the original files into a […]
Arcee AI recently released Trinity-Large-Thinking, a specialized artificial intelligence system designed to manage complex planning and automated workflows. The model generates visible reasoning steps before delivering its final answers, which […]
APEX is a new model compression method that reduces file size while preserving output accuracy. By adjusting precision levels based on specific layer and data roles, the technique matches high-fidelity […]
see-through automatically converts flat anime drawings into editable, multi-layered files suitable for animation. The system splits a single picture into up to twenty-three separate parts, like hair, eyes, and clothing, […]
ComfyUI-DreamScene360 is a new custom node that converts a single 360-degree panoramic image into a three-dimensional point cloud. The tool measures distances across the scene to build spatial layouts without […]
simple-captioner Version 1.0.2.1 now supports batch processing of images and videos through a standalone graphical interface. The updated tool automatically generates descriptive text files alongside media files located in user-selected […]
HybridScorer uses local GPU processing to quickly organize, score, and filter massive image collections. Version 1.6.5 quickly sorts visual assets into usable or discarded files while leaving originals untouched. Developer […]
Gen-Searcher is an open-source software tool that searches the web and gathers visual references before making new images. By running step-by-step lookups, it collects the exact facts and reference pictures […]
Adetailer-hires-sync is a lightweight utility that automatically enables face correction during high-resolution upsampling. It reverts the toggle to its previous state immediately after processing, eliminating manual clicks. KazeKaze93 developed this […]
PixlStash operates as a self-hosted image management platform designed to sort, filter, and evaluate extensive local photo libraries. The system runs entirely on your own hardware and provides a browser […]
The AgentScope team has released CoPaw-Flash-9B, a focused language model built to handle automated software tasks. It manages file operations, runs terminal commands, and tracks ongoing project states without constant […]
Microsoft has released harrier-oss-v1, a new family of multilingual text embedding models. These models convert text into dense mathematical representations to help computers understand meaning across different languages. This release […]
SycoFact is a lightweight four-billion-parameter model built to identify biased agreement and unsafe responses in large language outputs. The system scores replies across multiple safety categories while delivering a single […]
Llamafile delivers large language models through a single, portable file that runs locally without installation. This approach removes traditional setup steps and lets users start processing requests across Windows, macOS, […]
TagForge is an open-source application that streamlines the storage, editing, and analysis of image-text datasets for local machine learning workflows. It automatically pairs visual files with their text descriptions while […]
Unsloth Studio is a new web interface that lets users run and train AI models locally on their own hardware. The tool supports text, vision, and audio models across Windows, […]
Foundation-1 is a text-to-sample model built for structured music production. It generates tempo-synced, key-aware loops that slot directly into production workflows instead of producing generic audio textures. RoyalCities developed this […]
TurboQuant is a Python library that compresses high-dimensional vectors into compact 1-4 bit representations without needing calibration data or preprocessing. The tool uses a random rotation technique that transforms vectors […]
vmDeshpande's AI Agent Automation is a local-first workflow execution engine designed for AI-driven automation. Version 0.7.0 introduces dynamic decision-making capabilities, moving beyond simple step-by-step execution. The platform runs entirely on […]
ComfyUI-skill-public is a new open-source tool that connects an OpenClaw agent directly to ComfyUI for natural language control. Users can describe exactly what they want, and the system handles the […]
A new custom node for ComfyUI gives artists better control over image composition. Compose-Plugin-Comfyui lets users arrange multiple images on a canvas, resize them, rotate layers, and send the final […]
A new Slack app brings Hugging Face model tracking directly into team workspaces. The tool sends real-time notifications about model milestones, download counts, and organization activity straight to Slack channels. […]
GigaChat 3.1 Lightning is a compact language model built for fast local inference on consumer hardware. It uses a Mixture-of-Experts architecture with 10 billion total parameters, but only activates 1.8 […]
Granite-4.0-3B-Vision is a new vision-language model from IBM Research designed specifically for extracting structured data from documents, charts, and tables. The model converts visual information into machine-readable formats like CSV […]
ComfyUI-3D-Viewer-Pro is a new extension for ComfyUI that provides a professional-grade 3D model viewer and multi-pass rendering engine. Built with Three.js, this tool creates a seamless bridge between 3D assets […]
ComfyUI-Enhancement-Utils is a new custom node package that brings essential utility features to ComfyUI with full support for nested subgraphs. The toolkit includes resource monitoring, execution profiling, graph auto-arrangement, and […]
LumosX is a new framework for generating personalized videos with multiple subjects. It creates videos where specific people and objects stay consistent throughout, keeping the right attributes matched to the […]
Qwen3-TTS Easy Finetuning is an open-source tool that simplifies the process of training custom voice models. It provides a browser-based interface to manage the entire workflow, from processing raw audio […]
ComfyUI-Wan-VACE-Video-Joiner is a workflow tool that automatically stitches multiple video clips together while creating smooth transitions between them. The system uses VACE (Video-to-Video) technology to generate new frames at each […]
ComfyUI-Spectrum-WAN-Proper is a custom node for ComfyUI that speeds up WAN video generation. It uses a technique called Spectrum, which forecasts denoiser features instead of running the full network at […]
Matrix-Game-3.0 is an open-source interactive world model that generates real-time video at 720p resolution and 40 frames per second. It uses a memory-augmented architecture to maintain consistency over long video […]
Qwen-3.5-Abliterated-Comfyui-nvfp4 is a collection of quantized language models designed to function as AI assistants directly within ComfyUI. Developer Winnougan created these models to enable multimodal tasks like image analysis and […]
Z-Image-SAM-ControlNet is a new control model designed to transform segmented images into photorealistic pictures. It functions as a ControlNet for the Tongyi-MAI/Z-Image base model, allowing users to guide image generation […]
LongCat-AudioDiT is a new text-to-speech model that generates high-fidelity audio directly from text inputs. It operates directly on the waveform latent space rather than relying on intermediate acoustic representations like […]
ComfyUI-FBnodes is a collection of custom nodes for ComfyUI that streamlines video workflows and adds utility functions for AI content generation. The extension provides tools for video encoding with codec […]
World Model Bench is a new benchmark that tests whether AI world models can actually think about a scene rather than just generate smooth video. It measures cognitive intelligence through […]
PixelSmile is a new diffusion LoRA framework designed for fine-grained facial expression editing. It allows users to modify specific facial expressions in images with precise control over intensity levels, addressing […]
ComfyUI-YOLOE26 is a custom node pack that segments objects in images using text prompts. Users can type simple descriptions like "person," "car," or "red apple" to isolate objects without needing […]
Nemotron3-Nano-4B-Uncensored-HauhauCS-Aggressive is an uncensored version of NVIDIA's Nemotron-3 Nano 4B model. It removes built-in refusals and censorship mechanisms while preserving the original model's capabilities and personality. Developed by HauhauCS, who […]
ComfyUI-Darkroom is a custom node package for ComfyUI that provides professional color grading and film emulation capabilities. The suite includes 29 nodes covering film stocks, lens profiles, and comprehensive color […]
Sift is a desktop application designed to quickly organize large libraries of media files. It allows Windows users to sort images, videos, and audio files into specific folders using customizable […]
Toon-Tacular-Qwen-LoRA is a new LoRA model that brings the aesthetic of late 1990s and early 2000s cartoons to AI image generation. Created by renderartist, this tool was trained on 70 […]