ComfyUI-compatible Qwen 2.1 Turbo, from the official release: https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo

I did not make this in any way, just repackaged it for ComfyUI and quantized it.

Steps: 8

CFG: 1

Multiple quantizations available:

• BF16

• INT8 ConvRot (Needs ComfyUI v0.27.0+)

• W6A8 ConvRot (Needs ComfyUI v0.38.0+)

• W4A8 ConvRot (Needs ComfyUI v0.31.0+)

• NVFP4

They may not be in order in the list, refer to the filenames to be sure you are getting the right one.

In the additional files is a (very small) text encoder model, q3_k Qwen3 VL 8B, with vision tower intact but the lm_head removed. This means it cannot do prompt enhancement. It's gguf, so you'd need ComfyUI-GGUF or similar to load it in ComfyUI. Someone requested a text encoder smaller than 4gb. I tried q2_k but it was basically unusable.

Here's the example image above, but generated with the q3_k without the lm_head and the W4A8 model.