Animoo emo sprite making 101 via Stable Diffusion

Setup

  1. Pick a suitable anime model (very important for the booru tags!), eg. WD 1.4, AnyV3 etc..
  2. Optional, install the "tag complete prompt" extension
  3. Generate something

The example we'll be using
Image description

make sure the image is at least 512x512 minimum for best results

Inpainting

  1. Mask the face
    Image description
  2. Settings
    The most important setting to pay attention to is the "Inpaint masked" and the Denoising strength. Denoising strength would change depend what prompt you're using. Sometimes when you try a set of prompts, it might change the face to much, lowering it can help keeping it consistent.
    Image description
  3. Prompting
    You can actually delete all your prompts, with the exception any prompts that affects the face, in our case, I'll be leave the prompt, "crimson eyes" in.
    Then add in the emotion/facial expression you're trying to get. You can mix and match these to get the desired emo.
    Sometimes the changes are very subtle, you can increase the emphases of the prompt by (blush:1.3), if the changes are also messing up the face, you can also lower the denoising strength until it's normal
    Inversly, sometime when you combine two or more prompts, and one is overpowering the results you can deemphasize it (blush:0.8)
    Here are some I found that works well enough:

Example Prompts

Prompt Example
Happy Image description
Sad Image description
Smug Image description
Angry Image description
Scared Image description
Blush Image description
(Blush:1.3) Image description
(Ahegao:0.8) Image description
Tongue out Image description
Crying Image description
Wavy mouth Image description
Wink Image description
Closed eyes (remove any eyes related prompts for best results) Image description
Heart-shaped pupils Image description
Disgust Image description
Embarrassed Image description
Furrowed brow Image description
Confused Image description
Teeth Image description
Open mouth Image description
Closed mouth Image description
Frown Image description
Smile Image description
Sleepy Image description
Laughing Image description
Evil Smile Image description

Going further beyond

Sometimes the model just don't do your prompts. At this point I suggest opening up photoshop and drawing it out yourself, it doesn't need to be good, just needs the basic shape and color. Then just put the new image into inpainting and set the denoising to around ~0.5, mess around with it as needed. You could also use the inpaint sketch tab if you don't want to use an external painting program.

Edit
Pub: 09 Mar 2023 01:11 UTC
Edit: 09 Mar 2023 01:21 UTC
Views: 11179