Original post here

I run ComfyUI on my blisteringly slow RTX 3060 12GB locally and Perchance’s text-to-image-plugin absolutely blows it out of the water. The detail is INSANE. The intricate fractals made several people I asked – and even ChatGPT – think it’s Midjourney. I have not even done anything remotely CLOSE to this in ComfyUI.

So, what exactly is Perchance doing here? I don’t want “I think it’s xyz model” answers, I would like to see you all to try to replicate this image here as closely as possible. The exact settings I used for this are:

  • Positive prompt: A demonic god made of a swirling substance stands before a ruined city
  • Negative prompt is empty
  • No style
  • CFG: 7.0
  • Seed: 666
  • Image size: 512x768
  • I do not know what model this is, I have seen people claiming it is Chroma but I highly doubt it, as they have used Stable Diffusion 1.5(?) for a while (that model uses 512x768 in fact).

@perchance@lemmy.world I’d like to hear from you as to how I can achieve this locally, you’re the one who manages it after all :)

  • Koto@lemmy.worldEnglish
    1·
    2 months ago

    I’d also like to know. One thing is for sure, the base is flux.1-schnell with LoRas. Except, it still accepts negative prompts to some degrees (weakly), which flux shouldn’t do.

    • Iwaku@lemmy.worldOPEnglish
      2·
      2 months ago

      Schnell because of the speed I asume?

      Haven’t found good LoRAs to make that intricate style.

      Chroma does accept negative prompts! Even though it is somewhat based on Flux.

  • Danno@lemmy.worldEnglish
    1·
    2 months ago

    It’s beyond me for now, but maybe this helps: DIY Perchance API. The link is at 2/3rd of the Perchance welcome page.

    When it does work, let me know as I am in the process of getting myself new hardware esp. for LLM and the works. Any info on what and why will be highly appreciated.