Another attempt to do XL based checkpoint.

Based on Animagine4 Zero, merged with original trained lora. Yes, just one lora, because I don't know what else to do with it.

Prompting should be the same as all variant of Animagine4 since no changes made to the clip itself. You could check Cagliostro Lab's official guide, though this one has minor difference.

Also, please keep in mind that this is NOT A VPRED MODEL.

Quality prompt: masterpiece, absurdres

That's mainly it. You can try to add some other quality tag of Animagine4 but somehow they mess with image composition and I don't know why. Put at the end of your prompt.

Negatives prompt: none (or optional)

It's not really needed because they somehow can break image composition. Only use when necessary.

Usage:

Sampler: Euler A (Feel free to explore other samplers and schedulers)

Steps: 25-30 (You can try more than 30)

CFG: 5 (This is the recommended value from Cagliostro Lab, but you can try lower and/or higher)

VAE: Auto

Image size: Around 1MP is best. If you want to go higher (like 1024x1536), you'd better off doing upscale.

Upscaler (optional): R-ESRGAN4x + Anime6B (Feel free to explore other upscalers)

Many thanks to Cagliostro Lab for creating Animagine.