catboxanon

The current FreeU implementation comes from ComfyUI. The webui FreeU extension is a separate project. https://github.com/lllyasviel/stable-diffusion-webui-forge/blob/7b200553772b84c7a7fb3874f22a4af38a1448f7/extensions-builtin/sd_forge_freeu/scripts/forge_freeu.py#L4 https://github.com/comfyanonymous/ComfyUI/blob/ca7c310a0e8474c992b88fcf0f87040372f9c2c4/comfy_extras/nodes_freelunch.py#L64 Essentially what you're asking for is for the webui extension features to be...

Those 22 are just the LoRAs selected for their testbed. The paper mentions the LoRAs used are existing public ones. > [the ComposLoRA] testbed builds upon a collection of public...

Tried implementing this, results look worse though, so I'm going to close this as I don't see a point in supporting it. See https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/15037#issuecomment-2002655419 for an in-depth explantation.

The linked example shows an img2img task, which is already doable. The image provided for img2img is encoded to a latent by the VAE.

IA3 is not supported in Forge, because it is also not supported in ComfyUI (the LoRA backend Forge uses). https://github.com/comfyanonymous/ComfyUI/issues/1071#issuecomment-2019095689 I'm not sure where comfyanonymous "officially" mentioned this. Probably Matrix...

Will be fixed when https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/15035 is merged into this repo as part of whenever the next upstream sync happens.

> I also notice that the "Emphasis Mode" setting is also not correctly saved and load into the generated images. This would be related to upstream functionality, but I believe...

That's likely also fixed by the latter PR I linked.

FYI, soft inpainting doesn't work correctly with Forge at all currently. https://github.com/lllyasviel/stable-diffusion-webui-forge/pull/494

Ideally a PR should be made upstream. It should also be verified it doesn't break any functionality with non-RGBA images. https://github.com/AUTOMATIC1111/stable-diffusion-webui/pulls