Relationship with Kolors: Effective Training of Diffusion Model for Photorealistic Text-to-Image Synthesis

#8
by Maguro97 - opened

Hi, in the tech report I find the paper of "Kolors: Effective Training of Diffusion Model for Photorealistic Text-to-Image Synthesis".
The algorithm behind this project is based on the Kolors network (maybe fine-tuned)? Or is it a new architecture with only some elements in common?

Kwai-Kolors org

Yes, a fine-tuned version of Kolors with cloth image as reference

Does this training method strictly require paired data for finetuning? It seems a bit difficult to acquire the data.

Kwai-Kolors org

Yes, there is also open released paired data available, e.g., HD, DC virtual try-on datasets from https://paperswithcode.com/task/virtual-try-on.

Sign up or log in to comment