Qwen-Image-Edit-2511 In-Context Editing LoRA

This LoRA is released by DiffSynth-Studio and provides in-context editing capabilities for Qwen-Image-Edit-2511. You can use three images for the model: Image 1 as the base, Image 2 as the edited version of the base image, and Image 3 as the target image. The model will apply similar changes to Image 3.

Prompt: " Edit image 3 based on the transformation from image 1 to image 2 "

Personal experience based on tests of 10 images (5 SFW and 5 NSFW)

For NSFW you'll need a complementary LoRA. I used META4 at 0.8 and 0.86 strength.

Shift and LoRA strength play important roles here, especially in heavily modified images (Image 2).

Lower the weight to 0.86 if LoRA doesn't do well.

Shift between 1 and 3: the model will have more influence.

For a shift of 5 to 8, the model has less influence, and the LoRA weights (training data) and the prompt will have a greater impact.

These models are redistributed here for the sake of convenience.