Trending Model:#1Inklingthinkingmachines⬇16kTrending Model:#2Ternary-Bonsai-27B-ggufprism-ml⬇432kTrending Model:#3Bonsai-27B-ggufprism-ml⬇1405kTrending Model:#4Unlimited-OCRbaidu⬇2237kTrending Model:#5GLM-5.2zai-org⬇545kTrending Model:#6Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇2133kTrending Model:#7krea2-identity-editconradlocke⬇0kTrending Model:#8Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUFDavidAU⬇63kTrending Model:#9Qwen3.6-35B-A3B-Uncensored-HauhauCS-AggressiveHauhauCS⬇1998kTrending Model:#10OvisOCR2ATH-MaaS⬇17kTrending Model:#1Inklingthinkingmachines⬇16kTrending Model:#2Ternary-Bonsai-27B-ggufprism-ml⬇432kTrending Model:#3Bonsai-27B-ggufprism-ml⬇1405kTrending Model:#4Unlimited-OCRbaidu⬇2237kTrending Model:#5GLM-5.2zai-org⬇545kTrending Model:#6Qwythos-9B-Claude-Mythos-5-1M-GGUFempero-ai⬇2133kTrending Model:#7krea2-identity-editconradlocke⬇0kTrending Model:#8Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-NEO-MAX-MTP-GGUFDavidAU⬇63kTrending Model:#9Qwen3.6-35B-A3B-Uncensored-HauhauCS-AggressiveHauhauCS⬇1998kTrending Model:#10OvisOCR2ATH-MaaS⬇17k

SDMLX Supercharges Mac SDXL Workflows With MLX Native Speed

By vramkickedin | | 2 min read
Stylized MacBook with a glowing MLX chip on the screen.

SDMLX is a new alpha-stage ComfyUI node suite that runs SDXL natively through Apple’s MLX framework on Macs. It converts a checkpoint into a .sdmlx package once, then loads it fast while preserving familiar workflow patterns, and includes an early FLUX.1 MLX path. On an M1 Max, the developer measured SDXL workflows running about 25–30% faster than the standard PyTorch-MPS route.

Elef4nt-gh, a non-programmer using AI assistance, built SDMLX to make Mac image generation less of a compromise. The suite directly manages model data, memory, and speed patches through MLX instead of wrapping a CUDA-oriented toolkit. Complex workflows with ControlNet, IP-Adapter, and upscaling benefit most from this shift.

What works today

Key Features
  • One-click checkpoint conversion to .sdmlx format.
  • MLX-native sampling with image and latent output.
  • Built-in speed patch support (DMD2, Lightning, LCM, Hyper-SD).
  • LoRA loading with strength scheduling.
  • IP-Adapter, FaceID, and CLIP Vision paths.
  • Hires fix and tiled upscale nodes.

Creatives using ComfyUI on Apple Silicon Macs get a 25–30% speed boost for SDXL and a smoother multitasking experience. Privacy-focused professionals can run everything locally without the usual slowdowns that come from stacking IP-Adapter, ControlNet, and hires fix. The tool is especially helpful when combining multiple models in a single complex workflow.

Alpha status and upcoming plans

The developer warns that SDMLX is alpha: node names and defaults may still change. Currently unsupported are SD1.x, SD2.x, SD3, refiner checkpoints, and many non-SDXL architectures. Future plans include SEGS automation, SAM-based segmentation, and broader ControlNet support once the API stabilizes.

"In current local tests, SDMLX is typically about 25-30% faster than comparable PyTorch-MPS SDXL workflows on my M1-Max Mac Studio." — Source: Reddit

Project Links