Updated: Apr 17, 2026
styleDownload
1 variant available
The Anima Model is licensed by CircleStone Labs LLC. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.
Built on NVIDIA Cosmos
This is an experimental LoRA series trained on the Anima-preview model.
Version overview:
• data01-v1.0: Pure anime dataset, SDXL personally made mixed style dataset
• data02-v1.0: Another anime dataset, SDXL personally made mixed style dataset
• data03-v1.0: Based on data02 with additional partial 2.5D elements, can generate semi-realistic style with more sense of volume / depth appropriately
No specific trigger word required — load the LoRA and use descriptive prompts or Danbooru-style tags directly.
Common recommended settings (aligned with Anima official model card):
• Base Model → Anima
• LoRA Weight → 0.8–1.0 (0.8–0.9 usually most balanced)
• Sampler → er_sde / euler_a / Res Multistep
• Scheduler → Simple or SGM_Uniform
• Steps: 25–30
• CFG Scale: 4–5
Prompt tips:
- Start with quality tags like masterpiece, best quality, highres
- Negative prompt: In most cases, just add score_1, score_2, score_3 — this is often sufficient to avoid low-quality outputs. Adding more negative tags frequently impacts generation negatively (e.g., reduces variety or over-corrects details).
For more official Anima prompting details, check: https://huggingface.co/circlestone-labs/Anima