Image Wrangling

text2img, img2img, ...

Generation

AUTOMATIC1111
civitai.com has lots of fine-tuned models
Waifu Diffusion works best with anime and other cartoon art
- Danbooru style prompting

inpainting (demo):
- How to make any model into an inpainting model (Civitai discord post)
img2img models:
- depthmap2mask
- depth2img (better than depth2mask)
  - Tip: use Blender to create simple 3D scenes
- ControlNet ⭐
  - A1111 extension for ControlNet
  - Pose redactor (demo) for skeleton model
combined workflows:
- ComfyUI (examples)

Huh?
- What is textual inversion?
- What is dreambooth?
- LoRA vs Dreambooth vs Textual Inversion vs Hypernetworks video
- Textual Inversion = embeddings (~20kb) = new fine-tuned token = search for subject within existing model
- Hypernetwork = extra layers (different filesize) = a filter on top of another model
- Dreambooth = checkpoint (>2gb) or LoRA (different filesize) = model fine-tuned on instance token
- Fine-tuning = checkpoint (>2gb) = model fine-tuned on many tokens
- LoRA is both small and powerful so use that most of the time
Create and use:
- Textual Inversion:
  - Create and use embeddings in AUTOMATIC1111
- Hypernetwork:
  - Hypernetwork training for dummies
  - Tutorial Hypernetwork - Monkeypatch method
- DreamArtist (best one-shot?):
  - DreamArtist-sd-webui-extension
- Dreambooth:
- General fine-tuning:
  - EveryDream
Find:
- civitai.com has lots of embeddings and hypernetworks
- #ai-art-embeddings channel on Waifu Diffusion Discord
- Stable Diffusion Textual Inversion Embeddings

Topaz Mask AI ¹
Photobear ²
remove.bg
removal.ai
slazzer
Tip: some of these will only give you lower res versions in the free versions. You can use them as a mask on the original, to get original res.