Recommended usage

For best results follow recommended parameters:

• Checkpoint: AlbedoBase XL - v2.1

• CFG: 7-8

• Sampler: Euler ancestral

• Scheduler: Normal

• Image size: About 1024x1024. 720x1280, 1280x720 and similar

• Use natural language for prompting, not tags.

Caveats

• Training data does not include people only animals

• The training data (as of 0.1) includes text, which the model will try to replicate. Since this is built on the SDXL foundational model, its VAE and lack of text encoder prevents it from actually generating text, so it will be illegible.

Training data

• Physical book scanned and cropped: Dikkie Dik en de verdwenen knuffel. ( https://amzn.eu/d/9CYauee )

• Scanned pages were resized to 1024x1024