[IMPORTANT] Change the upstream to WebUI reForge.

[lshqqytiger]

I'm thinking of rebasing this repository on WebUI reForge.

What do you think? Let me know.

WebUI Forge
WebUI reForge

[TheFerumn]

What are difference between Forge and reForge ? I never tested reForge but honestly i was comparing lately ComfyUI and SD.Next to this AMD Forge fork and both work faster than Forge with DirectML.

[lshqqytiger]

I've never tried it, too. But people say it's faster and needs less vram as the original forge did before.

[VWiman]

As long as it doesn’t compromise ZLUDA support, I have no issue with making the change if it provides better performance for those using DirectML.

That said, I’m currently getting excellent results with ZLUDA, on a AMD iGPU.

[TheFerumn]

As long as it doesn’t compromise ZLUDA support, I have no issue with making the change if it provides better performance for those using DirectML.

That said, I’m currently getting excellent results with ZLUDA, on a AMD iGPU.

I guess you can quickly download reForge and compare speeds if you want but i can also suggest you to disable any automatic updates so you can always rollback to Forge based version if you really want.
@lshqqytiger Maybe just make reForge upstream version into another fork for testing ?

[TheFerumn]

So i took my time to test reForge and as i suspected it works way better with lora models. I found that for some reasons in Forge (this fork and original) i need to use this option for the Lora models to work properly

Without it i had different results than ComfyUI/SD.Next/reForge
In Forge without this setting faces in standard resolutions while generating full body are really deformed and it happens only while using Lora models. Its completely fine without them. Its not a big deal with ADetailer but still an issue for ppl who need to use inpainting in FP32.
Other than that there are no changes in generation speed but i am only testing DirectML and on really slow GPU so there are needed tests in ZLUDA with faster GPUs where difference in generation speed can actually be visible.