Thanks for open-sourcing this!
I can only find examples for 3 tasks in the examples/
directory: image_inpainting, depth_condition, and palette_condition. The remaining ones, like subject-driven, relighting, and instruction-based editing, etc. seem to be missing.
I tried subject-driven editing but got poor results—maybe the prompts used during training are different?
Show Example
System Prompt: You are an assistant designed to generate superior images with the highest degree of image-text alignment based on textual prompts and an object reference. <Prompt Start>
Prompt: A film-style shot. On the moon, this item drives across the lunar surface. In the background, Earth looms large in the foreground.
cond_pos_ids: offset
Input Image |
Output Image |
 |
 |