RSCV5 is an anime illustration style LoRA for Anima, focused on expressive eyes, soft blush, glossy hair highlights and shaded clothing. It can be used for portraits and multi-character scenes without artist-name tags.

Quick start

• Base model tested: Anima Base v1.0. This is an Anima LoRA; compatibility with other model families is not claimed.

• Trigger word: RSCV5

• LoRA strength: start at 0.9, applied globally to the model.

• Tested generation settings: 30 steps, CFG 4, Euler ancestral + SGM uniform.

• Tested sizes: 1024 × 1024 for portraits, 1152 × 768 for two characters, and 1536 × 1024 for three characters.

• Keep describing the characters, clothing, pose and background you want. Artist tags are not required.

Checkpoint

Final training checkpoint: 42,330 steps / 10 epochs, rank 32, alpha 16. A 15,000-step checkpoint was also retained for comparison; more training does not guarantee a better result for every prompt.

ComfyUI usage

Load the LoRA globally with the standard model-only LoRA loader. It works in a regular Anima workflow. For regional multi-character prompting and face refinement, I use my ComfyUI_RS_MultiCharacter_Sampler extension:

https://github.com/raspie10032/ComfyUI_RS_MultiCharacter_Sampler

Optional face refinement

Using RS MultiCharacter Detailer, I tested denoise 0.25, 20 steps, CFG 4, Euler ancestral + SGM uniform, SEGM face detection and WD matching for multi-character scenes. Start with a crop maximum size of 1024; higher processing resolutions are optional. Face-refined results should be distinguished from the initial generation when comparing the LoRA.

Limitations

This changes more than surface style: framing, pose and background can also shift. Results depend on the base model, prompt and seed. Character identity, exact composition and object counts are not guaranteed.

한국어

작가 태그 없이 사용하는 Anima용 스타일 LoRA입니다. 트리거는 RSCV5, 전역 강도는 0.9부터 시작해 주세요. 일반 Anima 워크플로에서도 사용할 수 있으며, 위 확장은 멀티캐릭터 영역 제어와 얼굴 보정에 선택적으로 사용합니다. 최종본은 42,330스텝이며, 그림마다 이전 체크포인트보다 항상 좋은 것은 아닙니다.