>>720037086
>I-it will keep getting better guise!
Yawn. I don't know about the LLM space but for the image gen stuff..
In the image generation side we're stuck with SDXL / Illustrious / PonyXL / NoobNAI for a year now. Flux is supposed to be the next step up but it's ridiculously demanding that using controlnets with it will just slow your rig down, it's not that much better than using SDXL's controlnet with basic 3d renders of the scene you want, depthmap it and then slowly start inpainting the various details (use relevant loras if you need to).
The only thing that people could 'improve' on is just finetuning (which is retardedly expensive, so only few people would bother). And then when they do release their own mixes of models, the rest of the prooompters would just complain it didn't get *their* waifu right, reupload those models somewhere else and such. If I were them I also don't wanna bother releasing free shit only to get complaints.
LLM side of things I guess it's only GPT5 or Qwen 3 of late but... Honestly... I don't really see as much value in LLM's if it doesn't have image recognition. I only use them as pseudo search engine and to 'OCR this content from this logbook into a table so I can copy paste it' or 'sum up the following values from a non-digital source' or 'translate these ching chong to english' lol. Basic tasks. Honestly just get it to be blazing fast for these simple tasks is enough for now.