[Change Request] Do not rescale image in TextEncodeQwenImageEditPlus
Open
- Dominant language
- Python
- Stars
- 133k
- Forks
- 15.7k
- Avg merge
- 1d 7h
- Merged PRs (30d)
- 158
Description
Qwen Image Edit 2509 comes with native support for ControlNet. When using it with a conditioning image of optimal resolution `1328x1328`, the conditioning image is downscaled to `1024x1024`. The impact is the generated result is downscaled and does not align with the conditioning image.
Example:
Depth Map Conditioning:
Current output (downscaled, the gate is smaller than the depth map):
Expected output (aligned with the depth map):
Would it be possible to NOT rescale the input image in `TextEncodeQwenImageEditPlus `? I think it should be users' responsibility to ensure the image has a proper scale before sending it to the text encoder.
Contributor guide
Assessment
This issue has not been assessed yet.