Hi all, first of all, I should mention that I’m pretty new to AI image generation.

I found this free image generator from perchance that’s really amazing – it covers many styles and produces very nice outputs. The only downside is the waiting time, which is understandable for a free tool. Now I’m wondering if I can somehow host something with similar results myself.

i read the image plugin page, which mention a 3GB download for the model. If anyone has a idea which model is used here, please let me know :)

I already have Comfyui up and running, messing around with SDLX, Flux and some other models.

Nothing i tried so far has nearly the same results.

I would appreciate any hint :)

  • Riskable@programming.devEnglish
    1·
    3 months ago

    Try the FLUX models and learn how to use that style of promting (it doesn’t use a negative prompt).

    My favorite right now is AnimePro FLUX which can generate great images in about 6 seconds on my 4060 Ti 16GB.

    • DasD@lemmy.worldOPEnglish
      1·
      3 months ago

      Thanks for your answer :) What do you mean by “that style of promting”? Woulnd i archive nearly the same results if i promt in the same style as in the offical perchance generator? Also it looks like the generator uses a wide variant of different loras and is capable to generate nearly everything from normal picture to nfws stuff.

      • Riskable@programming.devEnglish
        1·
        3 months ago

        FLUX uses a complete different style of prompting than Stable Diffusion. FLUX wants full sentences to describe what you want (sort of) whereas Stable Diffusion mostly wants keywords.

        Cool trick: You can ask ChatGPT or Gemini to convert a Stable Diffusion prompt to FLUX-style and vice versa to understand the difference. You can also give it a negative prompt and ask how you could incorporate that into a FLUX prompt 👍

  • Stein59@lemmy.worldEnglish
    1·
    3 months ago

    Yes, I am also very positively surprised by the AI img generator. And i too would like to know, what kind of program is behind all this and what models are used, and these style choices are they regular LoRA’s or what ? I too have used comfy and it has been so troublesome to get it consistent in style, and the style i want is realistic western cartoons, which isnt what the checkpoints usually come up with, but when i find something good, comfy may change to photo just because i add a word to the prompts. Such problems are not here. I am so happy. Cannot remember last time i turned off the ad block for anything. But its off here.