When using this LoRA, it is recommended to reduce the weights appropriately; otherwise, the generated image may exhibit severe oversaturation issues.

Bias, Risks, and Limitations

Relative to the base model, this style LoRA exhibits specific generative shifts when applied to prompts that specify dark‑skinned male subjects. An observable shift in rendered skin pigmentation is noted: the output tends toward a deeper tonal range than that produced by the base model under identical conditioning. Furthermore, within the same dark‑skinned demographic, a pronounced stylistic divergence is observed between male and female subjects—the generated female instances often present a visual aesthetic and textural quality that is markedly inconsistent with that of the male counterparts, to the extent that they appear to originate from distinct generative trajectories.

It should be emphasized that these observations are confined to the demographic groups predominantly represented in the training data. No systematic evaluation has been conducted for other skin‑tone or gender groups, and therefore no claims are made regarding the LoRA’s behavior or performance outside the described scope.

V4.5 U3

Anima model：anima-base-v1.0.safetensors

VAE：qwen_image_vae.safetensors

TE：qwen_3_06b_base.safetensors

res set:
bs20 512x512 3502images
bs6 1024x1024 3502images

V3 U3

Anima model：anima-base-v1.0.safetensors

VAE：qwen_image_vae.safetensors

TE：qwen_3_06b_base.safetensors

res set:
bs8 1024x1024 1181images

License & Usage Notice

This LoRA model is a derivative work based on the anima-preview model by CircleStone Labs, and is therefore governed by the following licensing terms:

Key Points:

1. Non-Commercial Restriction on the Model: The anima-preview base model and any derivatives (including this LoRA) are licensed under the CircleStone Labs Non-Commercial License. Commercial use of the model itself is strictly prohibited. This means you cannot:

◦ Host this model (or derivatives) on a platform and charge for access/generation.

◦ Use it behind a paid API.

◦ Integrate it directly into a paid product (e.g., a game, software).

2. Commercial Use of Outputs is Allowed: You may use images generated with this model (the "Outputs") for commercial purposes. Examples include:

◦ Creating art for visual novels, video games, or anime.

◦ Generating concept art, illustrations, or assets for commercial projects.

◦ Using generated images in advertisements.

3. Derivative Works & Distribution: This LoRA is explicitly considered a Derivative Work under the license. The license is designed to be permissive for end-users and derivative creators:

◦ The licensor (CircleStone Labs) does not have a "rugpull" clause and cannot arbitrarily terminate the license or demand takedowns of compliant derivatives.

◦ Commercial licensees (e.g., platforms) may be granted rights to use derivatives like this LoRA.

4. Upstream License Inheritance: The anima-preview model is itself a Derivative Model of Cosmos-Predict2-2B-Text2Image, and thus remains subject to the relevant terms of the NVIDIA Open Model License Agreement.

5. Legal Compliance: Users are responsible for ensuring their use of the model complies with all applicable laws. The model creator/licensor is not liable for misuse.

Summary for Users:

✅ You CAN: Use this LoRA for personal or research purposes. Use the images it generates commercially.

❌ You CANNOT: Use this LoRA model itself for any commercial service or product.

⚠️ You MUST: Comply with applicable laws and the terms of both the CircleStone Labs Non-Commercial License and the upstream NVIDIA Open Model License.

License Links:

• CircleStone Labs Non-Commercial License (Primary): https://huggingface.co/circlestone-labs/Anima/raw/main/LICENSE.md

• NVIDIA Open Model License Agreement (Upstream): https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/

• Base Model: anima-preview on Hugging Face

By using this model, you agree to be bound by these terms.

Built on NVIDIA Cosmos