# low VRAM QwenImage2.1 INT4 & INT8 ConvRot — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/b24acfa289981e1c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-09-20 15:50 ingested 2026-09-21 02:37

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1wlklfj/low_vram_qwenimage21_int4_int8_convrot/)

## retained text



```
Cat pic: default/ComfyUI T2I workflow,  
toxicdog's Int8ConvRot  
clip model: w4a8  
Prompt: `GrumpyCat looking at camera.`  So I'd say good prompt adhesion, though I was hoping the model would have ESP & knew I wanted THE "Grumpy Cat" which has same face shape but dark & light gray.

INT8 fits 8GB VRAM, INT4 fits 4GB.  
[https://huggingface.co/toxicdog/Qwen-Image-2.1-INT4ConvRot-ComfyUI](https://huggingface.co/toxicdog/Qwen-Image-2.1-INT4ConvRot-ComfyUI)
```



document fetched 2026-09-21 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1wlklfj/low_vram_qwenimage21_int4_int8_convrot
