Great Job! but would it support SD 1.5 models? about fooocus-mre HOT 8 CLOSED

@xy185 Fooocus was designed as UI exclusively for SDXL, so I think it just refused to load non-SDXL model you've selected, and image was probably generated using last SDXL model you've picked. You can check the log file (log.html in output folder) to see what model was used to generate particular image.

from fooocus-mre.

Theoretically it could be possible to support SD (comfy backend used by Fooocus supports SD models), but it would complicate the UI - older SD models works best using 512x512 (1.5) or 768x768 (2.1) as base resolutions, for example. So... while I still enjoy and sometime use older SD models, for now I'd rather focus on main purpose of Fooocus, that is being solid UI for the SDXL.

And as for anime fairies in Fooocus - you can try sai-anime style, and SDXL variants trained on anime, like CounterfeitXL :).

from fooocus-mre.

Theoretically it could be possible to support SD (comfy backend used by Fooocus supports SD models), but it would complicate the UI - older SD models works best using 512x512 (1.5) or 768x768 (2.1) as base resolutions, for example. So... while I still enjoy and sometime use older SD models, for now I'd rather focus on main purpose of Fooocus, that is being solid UI for the SDXL.

And as for anime fairies in Fooocus - you can try sai-anime style, and SDXL variants trained on anime, like CounterfeitXL :).

@MoonRide303 Thanks so much for your reply!
I was not so familiar with XL, thanks for your kindly guide, and I would try the style and new model,

Finallly, thanks for your great work: Fooocus-MRE, it's perfect!

from fooocus-mre.

Limited support for SD models added in v2.0.78.3 (commit 6d47547).

Test image generated using DreamShaper 8:

Not supported functions: refiner, Control-LoRAs, Revision, inpainting, outpainting. You need to manually pick compatible resolution and LoRAs, too.

from fooocus-mre.

@Aamir3d Yup. Inpainting in Fooocus is based on supplementary model designed for SDXL (custom type of LoRA) - there is no easy way to make it work with non-SDXL models. Inpainting with SD in general is doable (using dedicated inpainting models), but I would need to change current workflow quite a bit to make it work. Image-2-Image should work fine.

from fooocus-mre.

One of the reasons to have a 1.5 model support is the speed of generation and resource consumption. SDXL doesn't play very well with systems that have 4/6GB VRAM and 16GB RAM.

For example, I can generate an image at 768 x something in A1111 in less than 10 seconds using only 4 GB of VRAM in a laptop GPU.
For SDXL, the same generation (1024+ res) in my laptop GPU takes over 2 minutes because of memory requirements with Fooocus-MRE (tried this just now).

It's a different matter on my desktop, where I've got a 12GB GPU and 32 GB RAM, SDXL is much better in that scenario.

from fooocus-mre.

@Aamir3d Yeah, that's true - SDXL isn't really usable on GPUs below 8 GB VRAM.

from fooocus-mre.

Thank you @MoonRide303 ! So, image2image will still work, but not inpaint?

from fooocus-mre.