# Qwen Image 2.1 turbo int8 workflows: Text to image and Edit — retained record

> Retained reddit record from reddit:comfyui.

Canonical: https://highsignal.app/data/records/d9fd4513f7888382

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:comfyui)  published 2026-10-13 12:40 ingested 2026-10-14 02:42

[original source ↗](https://www.reddit.com/r/comfyui/comments/1x4vjt5/qwen_image_21_turbo_int8_workflows_text_to_image/)

## retained text



```
To build these workflows, you can download the Turbo INT8 convrot version of Qwen Image 2.1 on [HuggingFace](https://huggingface.co/Comfy-Org/Qwen-Image-2.1/tree/main/diffusion_models) and add it to your diffusion\_models folder. Change the UNet name and change the number of steps to 8 in the default wokflow.

Performance (1024x1024) Text to image: \~4 seconds for a second-generation image on an RTX 5070 Ti; \~8 seconds on an RTX 5060 Ti 16GB.

Performance (1024x1024) Edit: \~7 seconds for a second-generation image on an RTX 5070 Ti; \~14 seconds on an RTX 5060 Ti 16GB.

These workflows are on [Civitai](https://civitai.com/user/def813/models?sort=Highest+Rated&tag=base+model) and [Github](https://github.com/H2logs/Qwen-Image-2.1-turbo-int8-workflow).
```



document fetched 2026-10-14 00:42 · canonical https://reddit.com/r/comfyui/comments/1x4vjt5/qwen_image_21_turbo_int8_workflows_text_to_image
