Does anyone else can't stand ComfyUI and prefers classic Automatic/Forge UI or it's just me?

I tried. I swear I tried.

Each time a new "ultra super mega" model appears I'm reinstalling that bloody UI just so I can test it, and each time some God Awful error list appears telling me that somehow the latest version it's missing something critical that needs to be downloaded ASAP. And in 3/4 of cases that's not the right download, it's incompatible or needs additional stuff to work.



Even when everything works out of the box I am looking at the final image and think to myself "I could have had 4 batches of 8 images to choose from instead of 1-4 like here".

I am not lazy, but I have a job and I don't have hours to spend to make sure each new model and each new workflow that I download is compatible with every single other version of all dependencies.


Does anyone else feels the same or I'm just the only one that keeps using Forge and Automatic?...

https://redd.it/1tt5uiq
@rStableDiffusion

Damn Anima Base is cooking! What's your favorite lora?
https://redd.it/1tt6eir
@rStableDiffusion

ComfyUI_HYWorld2 update. Quality improvement + World Stereo Light models!

https://redd.it/1tt61vq
@rStableDiffusion

PIT NVIDIA vs SeedVR2

https://redd.it/1tt8h2w
@rStableDiffusion

Perceptual LoRA Toolkit now supports Z-Image Turbo

https://redd.it/1ttak8l
@rStableDiffusion

lora dataset images and captions

Okay. I hear a lot of do and don't's, but *gawd* damn, I need more.

Character lora. 25 images. All 1024x1024, all consistent, varying, ...in my mind complete for at least a *functional* if not *flexible* lora.

How the hell do I caption this to be easy model side? I dont want to have to fine tune knobs and prompt engineer like Gemini and other llms are doing to my captions.

I have a highly toxic and inflexible lora iteration right now, I'm not dumb enough to require crash coursing, but im stuck.

I know the *"transient state"* of the image as a whole, including viewpoint should be tagged, but how does one ensure accuracy for the training of the character?

TriggerWord, camera angle, objects/lighting/background to not triggerword bake, but what ELSE about the character needs to be captioned for flexibility in the character themselves? I know clothing and accessories so a bunch of crap doesnt get welded to the character, but hairstyles and expressions?

Those *make* the character, but doesn't tagging them.... remove them from the character? .....but then dont all expressions and hairstyles get averaged and welded together?

https://redd.it/1ttdbmo
@rStableDiffusion

Best anime model for multiple characters or Lora?

Hey everyone

I'm really struggling with multiple characters and their details. I can do a prompt that says 2girls but I sometimes get 3, etc. Or try specifying characters and their details and they'll be opposite Ori say I want small breasts and I get huge ones or something random. Or I get the perfect prompt and I regenerate it or read n later and it'll be f*cked. Is there something I can do/use in the prompt or a Lora I can use or something? I've tried pony, illumi, illustrious, noobai, wai-illustrious

Hope you can help

Regards

https://redd.it/1tt95u3
@rStableDiffusion

FLUX.2-klein-base-9B ControlLight LoRA Release for changing lighting of a photo
https://yfyang007.github.io/ControlLight/

https://redd.it/1ttfv5z
@rStableDiffusion

Nvidia releases Cosmos3-Super-Image2Video . 64B parametres
https://redd.it/1ttjuc0
@rStableDiffusion

Nvidia releasesCosmos3-Super-Text2Image model . 64 billion paramteres
https://redd.it/1ttjrip
@rStableDiffusion

The Cosmos omnimodel family of models - 3 variants Edge(4B) , Nano(16B) , Super (64B)

https://redd.it/1ttka77
@rStableDiffusion