Deaquay Streamlines Workflows With ComfyUI-Qwen3.5-Uncensored

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 use.

Developer Deaquay created this extension to fill a gap in current node libraries, specifically targeting prompt generation tasks that require uncensored or modified weights. The project supports standard and GGUF formats while respecting ComfyUI directory structures.

Automatic local model discovery

• Scans local directories for GGUF and HuggingFace weights automatically.
• Detects model architecture from file metadata instead of relying on file names.
• Pairs mmproj vision files with matching models without extra configuration.
• Includes a lightweight GGUF-only branch to bypass Transformers dependencies.
• Supports standard, abliterated, and community-modified model variants.

Operators running text pipelines will save setup time since the system recognizes finetuned checkpoints the moment they are placed in the designated folders. Users managing batch image scripts can integrate custom vision weights directly into existing graphs.

Developer notes and setup

The creator built this extension after finding existing libraries unresponsive to community feedback. Installation requires specific dependency handling, including downloading llama-cpp-python from a community fork rather than standard package managers. Transformers must also be updated to version 5.2.0 or higher before operation.

Deployment relies on manual CUDA environment verification.

"Wanted to use qwen3.5 for autoprompts, but devs are sleeping so I patched QwenVL-Mod to support it,"

explained the developer in a community post. Operators planning to run larger quantized weights should verify their hardware matches the selected model size.

Download ComfyUI-Qwen3.5-Uncensored from GitHub.