Request to implement FreeScale, a new diffusion scheduler

[arthur-qiu]

Model/Pipeline/Scheduler description

FreeScale is a tuning-free method for higher-resolution visual generation, unlocking the 8k image generation for pre-trained SDXL! Compared to direct inference by SDXL, FreeScale brings negligible additional memory and time costs.

Open source status

• The model implementation is available.
• The model weights are available (Only relevant if addition is not a scheduler).

Provide useful links for the implementation

• Project: http://haonanqiu.com/projects/FreeScale.html
• Paper: https://arxiv.org/abs/2412.09626
• Code: https://github.com/ali-vilab/FreeScale
• Hugging Face Demo: https://huggingface.co/spaces/MoonQiu/FreeScale

The code changes of FreeScale are not complicated, but I do not know how to integrate them into diffusers smoothly. If you have questions about FreeScale, please ask me(@arthur-qiu).