CoolFace
Modelpublic

ij/anima-humanize-composition-r32-v1

sourceHugging Faceotherupdated 3mo agoView on Hugging Face
0likes52downloads
Model Card

Anima Humanize Composition LoRA r32 v1

Anima Humanize Composition LoRA is an experimental LoRA for circlestone-labs/Anima.

It is intended to gently bias Anima toward more cover-like character placement: clearer subject focus, less awkward cropping, and more deliberate page composition.

This is not a general quality LoRA, character LoRA, artist LoRA, or prompt replacement. Use it as a small composition bias on top of your usual Anima workflow.

[image]

Model File

text
anima_humanize_composition_r32_v1.safetensors

Recommended Use

Start low.

text
LoRA strength: 0.35 - 0.55

In quick tests, 0.5 already has a visible effect. 1.0 can be too strong and may change framing, scene content, or character emphasis more than expected.

Suggested positive tags:

text
novel cover, good composition

Suggested negative tags:

text
expanded view, bad composition, cropped, bad crop, ai generated

Example ComfyUI stack:

text
Base model: Anima
Turbo LoRA: optional, if your Anima workflow already uses it
This LoRA: 0.35 - 0.55

What To Expect

The LoRA tends to:

  • —Pull the main subject into a clearer cover-style focal position.
  • —Reduce overly distant or weakly staged character placement.
  • —Favor stronger character focus and more intentional framing.

It may also:

  • —Make the subject larger than desired at high strength.
  • —Push prompts toward a more polished cover illustration look.
  • —Alter scene details when used too strongly.

Comparison

The image below compares a baseline run, an intermediate checkpoint, and the final LoRA in the same evaluation setup.

[image]

Training Data

The training data details are intentionally not disclosed. The LoRA was trained on a private curated dataset for composition and framing behavior.

License

This LoRA is a derivative work of Anima and follows the licensing constraints of the base model. Please refer to the Anima model page and its CircleStone Labs non-commercial license before use.

This model is intended for research and non-commercial personal experimentation.

Disclaimer

This is an experimental composition-control LoRA. It does not guarantee better results for every prompt, seed, aspect ratio, or sampler. For best results, test several strengths and keep the LoRA weight modest.