A Community Krea 2 LoRA for Anime Characters Hits 864K Downloads
Krea2_Asian_Character LoRA trends on Hugging Face with roughly 864K downloads on the model page . Trained on Krea 2 Raw using AI-Toolkit, targeting Japanese character styles with varied framing. Works on both Raw and Turbo checkpoints, following Krea's train-on-Raw, run-on-Turbo workflow.
Takeaways
- Krea2_Asian_Character LoRA trends on Hugging Face with roughly 864K downloads on the model page .
- Trained on Krea 2 Raw using AI-Toolkit, targeting Japanese character styles with varied framing.
- Works on both Raw and Turbo checkpoints, following Krea's train-on-Raw, run-on-Turbo workflow.
- Recommended weight 0.8 to 1.0, with 20 to 30 inference steps for best results.
- Base model is a 12B Diffusion Transformer with native ComfyUI support .
- Released under MIT, but downstream use must respect Krea 2 Community License on the base.
A community LoRA for Japanese-style character generation has entered Hugging Face’s trending ranks, with an external estimate placing downloads near 864,000. Published by user fidel1234, Krea2_Asian_Character specializes Krea AI’s open-weight image model and supports both Krea 2 Raw and Krea 2 Turbo, according to its model card. Hugging Face does not provide a native download count for this repository, so the figure remains an estimate.
The release gives developers a concrete way to evaluate Krea’s two-checkpoint strategy. Raw provides a flexible base for training, and Turbo uses distillation to generate images in eight inference steps. An adapter that transfers cleanly between them can shorten the path from fine-tuning to interactive or production inference.
A compact style layer
A LoRA stores a small set of weight adjustments rather than a complete model checkpoint. Generation tools apply those adjustments to the base model at inference time, reducing storage and training costs while adding a specific style, subject, or behavior.
The creator trained this adapter on Krea 2 Raw with AI-Toolkit, according to the model card . Its hand-curated dataset includes close-up, medium, full-body, and varied-angle images intended to support several compositions. The card lists the following starting points:
| Setting | Published guidance |
|---|---|
| Training base | Krea 2 Raw, trained with AI-Toolkit |
| Compatible checkpoints |
Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.
Read original report
