Repository navigation
If Flux is not supported, the Stable Diffusion webui project can be abandoned. Novices should not waste their time learning on this project. #16476
Description
If Flux is not supported, the Stable Diffusion webui project can be abandoned. Novices should not waste their time learning on this project.
Reacted by nitinmukesh
Reacted by dill-shower, Yellowcooln, ChrisWhite, Fancy2424, qinlige, highnrgappalachian, dorkulon, maudslice and Won-Kyu Park
Activity
[-][Feature Request]: [/-][+]If Flux is not supported, the Stable Diffusion webui project can be abandoned. Novices should not waste their time learning on this project.[/+]mnlkrp commentedon Sep 10, 2024
Atm FLUX has no LORAs, no finetuned models, no ControlNet, no anything basically.
Meanwhile SD1.5 and SDXL has it all. So no, Automatic1111 is still one of the best Stable Diffusion tools.
skyliwq commentedon Sep 10, 2024
It’s just a matter of time.
It’s just a matter of time, and moreover, there are other UIs as well
dill-shower commentedon Sep 10, 2024
This isn't a matter of time; it's about the fundamental limitations of the FLUX architecture, intentionally implemented by the developers. What they sold you as open-source and supposedly better than Stable Diffusion isn't actually that in practice. The only difference between FLUX and closed models like DALL-E is that FLUX uses your hardware for image generation instead of the company's servers.
All FLUX models you can find are distilled versions. Any LoRA created for them adds something but distorts the overall picture. Fine-tuning is impossible; all attempts to fine-tune the model so far have only resulted in quality degradation. This will continue until the developers release the original Pro model to the public. While StabilityAI has promised to eventually release their largest models as open-source, FLUX developers have never made such commitments. FLUX Pro, which is the original and most effective model, is only available through API. We can only dream that someday they'll cater to their fans and allow them to create their own models. For now, you can create LoRAs and train them for specific purposes, but this severely compromises the model's general capabilities. Your so-called open FLUX model is questionably open at best.
highnrgappalachian commentedon Sep 11, 2024
You had this cringe take and posted it as an issue? Seriously?
skyliwq commentedon Sep 12, 2024
This is not about whether weui and flex are easy to use. We need a comprehensive UI that can incorporate the latest technologies, such as Comfy UI and Webui Forge
nitinmukesh commentedon Sep 12, 2024
I agree with OP. I recently moved to Forge.
Shut this project and archive it. Join the team at Forge and support it. More the better we will get new features in Forge fast. Forge is anyway Auto1111 WebUI but with support for newer models.
E-Studios commentedon Sep 12, 2024
I agree flux isnt the checkpoint we are hoped for... out of the box it doesnt even work well... people say hands and feet work perfectly... no it aint. yes Flux looks great with realistic images but i think most used and most compatible is SD1.5/SDXL and Pony.
Anyways... yea A1111 really needs put more people together because its getting outdated each hour meanwhile ForgeUI is faster and supports even more stuff.
Forge is just like A1111 but better. (still using A1111) atm for most of the work because im lazy to set it up.
dill-shower commentedon Sep 12, 2024
You can make a pull request and if it doesn't break anything, it will be accepted. But spending effort on supporting a hopeless model for which the developers decided not to even publish documentation is wrong. Even sd3, which was criticized so much on reddit, has more prospects and its support in webui is not ideal.