Comments

Blacklisted
    (help)
    Show Search Options
    post #639
    77 133 4 E

    FurrAI

    Member

    kemowolf said:
    How is this AI?!

    It's actually two images. As far as I know, there's no way to specify multiple character trainings with AI. Each character was rendered separately, combined into one image, and run through image-to-image with a low strength to make the style match better. The text was added by hand.

  • |
  • 4
    •
    post #1008
    148 229 2 E
    post #639
    77 133 4 E
    post #957
    1 3 1 Q

    ShinyLuxio

    Member

    If you don't have dedicated graphics card, I suggest you to use Google Colab environment. You can even choose to use NovelAI (easier to use, less customizable) or Stable Diffusion (harder, but you can customize everything)

  • |
  • 2
    •
    post #922
    41 64 2 E
    post #940
    10 6 1 S
    post #806
    8 9 2 S
    post #922
    41 64 2 E
    post #389
    9 4 1 E
    post #806
    8 9 2 S

    GexoAI

    Member

    Those are some... very specific negative prompts. Good thing you added them too, because nobody needs to see that!

  • |
  • 1
    •
    post #787
    3 13 3 E

    agreenfish

    Member

    gotadraw said:
    Yeah, they do haha. I don't edit any of these pictures once I generate them, I just keep them in raw form.

    if you're generating these with the Automatic1111 UI, the head is an easy fix - do inpainting on it with the "Inpaint at full resolution" box fixed, source original with maybe 8px mask blur. Drop the noise a bit, maybe to 0.5, run Euler a at maybe 30 or 40 iterations. That should clean it up; use the batch size option to select from the best alternative.

  • |
  • 0
    •
    post #796
    69 87 1 E
    post #787
    3 13 3 E

    GotADraw

    Member

    kyiiel said:
    The eyes and hands need work.

    Yeah, they do haha. I don't edit any of these pictures once I generate them, I just keep them in raw form.

  • |
  • 0
    •
    post #787
    3 13 3 E
    post #731
    42 72 1 E
    post #701
    30 25 1 E
    post #639
    77 133 4 E
    post #644
    94 155 2 E
    10 18 1 S
    post #370
    1 5 1 E

    director-mo

    Member

    Is rope_bondage supposed to look like a noose? lol (I also regret looking up the e621 wiki page for a noose, the pictures were less playful than I expected. Dead dove meme... Almost literally.)

  • |
  • 0
    •
    post #243
    1 7 2 S
    post #61
    8 13 1 S
    post #74
    61 123 1 E
    post #239
    4 9 2 S

    Gem

    Member

    lance_armstrong said:
    A lot of the AI images uploaded so far seem to have a very high contrast between dark and bright areas. Not necessarily from you, Gem.

    Also, shouldn't it be possible to get rid of all attempted signatures with a little inpainting?

    Very contrast heavy and yes, with inpainting or smudging in photoshop/similar

  • |
  • 1
    •
    post #239
    4 9 2 S

    Lance Armstrong

    Former Staff

    A lot of the AI images uploaded so far seem to have a very high contrast between dark and bright areas. Not necessarily from you, Gem.

    Also, shouldn't it be possible to get rid of all attempted signatures with a little inpainting?

  • |
  • 1
    •
    post #243
    1 7 2 S
    post #184
    12 14 2 S
    post #184
    12 14 2 S
    post #3
    130 124 51 S
    post #12
    9 8 1 S
    post #3
    130 124 51 S
    post #3
    130 124 51 S