AI BriefingPickQwen-Image-2.1-Text-Encoder-Heretic-GGUF: Fix Qwen-Image-2.1 GGUF Loading Issues with ComfyUI-GGUF-Qwen3VL-TE
pottokao/Qwen-Image-2.1-Text-Encoder-Heretic-GGUF
About the project
The text encoder for Qwen-Image-2.1 is provided in three formats: GGUF, FP8, and bf16. It includes support for a dedicated node (ComfyUI-GGUF-Qwen3VL-TE) that resolves the 'input size' error caused by ComfyUI-GGUF failing to load the Qwen3-VL vision tower. Placing the mmproj file in the same folder as the GGUF file resolves the issue without requiring a re-download.
You can select optimized files based on your hardware environment. For minimal capacity, the 5.0GB Q4_K_M GGUF and 1.2GB mmproj files are recommended. For NVIDIA GPU environments, the 9.3GB FP8 safetensors file is recommended. For non-CUDA environments such as Mac, use the 17.5GB bf16 file.
Since Qwen-Image-2.1 is a non-distilled model, 25-step generation is recommended. For image generation without text, setting cfg to 1.0 yields the smoothest and fastest results. It is distributed under the Apache 2.0 license, and the abliterated (safety filter removed) version allows for more flexible prompt processing.
pottokao/Qwen-Image-2.1-Text-Encoder-Heretic-GGUF
The original page has no description.
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.