Skip to content

[Prompt Diffusion] In-Context Learning Unlocked for Diffusion Models #6214

@sayakpaul

Description

@sayakpaul

In short, prompt diffusion allows for in-context learning for diffusion models.

Code: https://github.com/Zhendong-Wang/Prompt-Diffusion
Models: https://huggingface.co/zhendongw/prompt-diffusion

One should be able to use https://github.com/huggingface/diffusers/blob/main/scripts/convert_original_controlnet_to_diffusers.py to get the model converted into diffusers format.

input_cond_block seems to be the new addition: https://github.com/Zhendong-Wang/Prompt-Diffusion/blob/d5737b81497c17af8d1e0b242f03dadab93bec1c/cldm/cldm.py#L165. This needs to be incorporated in the conversion and in the ControlNet block.

And then it's incorporated like so:
https://github.com/Zhendong-Wang/Prompt-Diffusion/blob/d5737b81497c17af8d1e0b242f03dadab93bec1c/cldm/cldm.py#L302 (execution starts from https://github.com/Zhendong-Wang/Prompt-Diffusion/blob/d5737b81497c17af8d1e0b242f03dadab93bec1c/run_prompt_diffusion.ipynb)

Happy to help anyone willing to take this one up!

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions