Qwen-Image 2.1 Turbo INT4 ConvRot
Repackaged model files for ComfyUI and others, with INT4 ConvRot converted for testing in Radiant Canvas for fast macOS inference: Radiant Canvas on the App Store.
The original model is Qwen/Qwen-Image-2.1-Turbo, the 8-step accelerated checkpoint of Qwen-Image-2.1. The ComfyUI single-file BF16 transformer used for this conversion came from Comfy-Org/Qwen-Image-2.1.
File
diffusion_models/qwen_image_2.1_turbo_int4_convrot.safetensors: the Turbo diffusion transformer in ComfyUI's native ConvRot W4A4 format.
For ComfyUI, place the file in:
ComfyUI/models/diffusion_models/
Use a recent ComfyUI build with Qwen-Image-2.1 and native INT4 ConvRot support, and load it with the standard diffusion model loader. The text encoder and VAE are not included here. Use the Qwen-Image-2.1 ones from Comfy-Org/Qwen-Image-2.1 or toxicdog/Qwen-Image-2.1-INT4ConvRot-ComfyUI, which also has an INT4 ConvRot text encoder.
Turbo is meant for 8 steps at CFG 1.
Conversion details
- Original model revision:
d65dbc9a7e8f6b5479e33dee6030eaab2a906509 - Comfy-Org repack revision:
72a49122613a4f9afbcc443426d6c765c0c383cf - Source:
qwen_image_2.1_turbo_bf16.safetensors, 265 tensors - Quantized layers: 193 (attention Q/K/V/out and image MLP in all 32 blocks, plus the shared modulation), the same layer set as the base Qwen-Image-2.1 INT4 ConvRot file
- ConvRot group size: 256
- Quantization group size: 64
- Output size: 3,666,512,176 bytes
- SHA-256:
1ea2e40d3cdac4eb06a05bcd0dbeaf67f131ba01d3ed0519708ca61a6efed29e
Text input, timestep embedding, image input, output projection and normalization layers are kept in BF16.
This is a test conversion. Check output quality for your own prompts and workflows before relying on it.
License
The original model is released under the Qwen Research License Agreement. A copy of the license is included in this repository.
- Downloads last month
- 788