FWAI ANIMA Magic

Support & Official Links

Note: If you enjoy my model content, please support me with a donation on my Kofi page. Your support helps me continue developing tools and models for everyone. Download links for the models will be available on my Kofi page for users who support the development of my models, allowing them to be used locally. Thank you for your attention.

https://ko-fi.com/jaceraimi

This model is officially presented by Jace Raimi on the following pages:
https://tensorhub.art/models/992095100488169536
https://tensor.art/models/992095100488169536
https://www.seaart.ai/es/models/detail/d78ivide878c739gkp8g
https://tungsten.run/model/7nwHdDucqa?model_version=Daoixf7xcY
https://yodayo.com/models/06aa64d1-5e56-49cb-b0b3-096b21bf2d07?index=1&modelversion=fb37b6a9-c107-412e-96a4-c1bf878d965c
https://moescape.ai/models/06aa64d1-5e56-49cb-b0b3-096b21bf2d07?index=1&modelversion=fb37b6a9-c107-412e-96a4-c1bf878d965c


📜 Changelog — FWAI_ANIMA_Magic_v1.8.6_(R2)

🎨 Style and Tags

    Enhanced anime style: more detailed, screen-cap quality, with fine-grained fidelity.

    Improved integration of Danbooru-style tags, enabling richer and more precise prompt control.

    Better recognition of popular anime characters, with more coherent poses and expressions.

    More natural anatomy and consistent proportions across outputs.

✨ Quality Functions & Style Triggers

    Implemented activation triggers for high-quality rendering: 
    
    masterpiece, best_quality, absurdres, anime_coloring, 2D, anime_screencap, official_artstyle, traditional_media, cinematic_lighting.
    

    Added support for speech bubbles and signs in prompts:

        She has a speech bubble that says "insert text here."

        He has a speech bubble that says "insert text here."

        She's holding a rectangular sign out in front of her that reads "FWAI ANIMA MAGIC".

        He held a rectangular sign in front of him that read "FWAI ANIMA MAGIC".
        

    Extended resolution support up to 2000 × 2000, enabling 1080p outputs.

        Note: This feature is currently functional only in ComfyUI, as web platforms remain limited to Anima’s default resolutions.
        

    Expanded Danbooru tag coverage for more natural English and prompt triggers, similar to Illustrious workflows.

🛠️ Technical Improvements

    Optimized inference for prompts with multiple tags.

    Reduced artifacts in complex backgrounds and dynamic lighting scenes.

    Improved balance between sharpness and smoothness in anime-style renders.

    Increased stability for long prompts and advanced tag combinations.

⚙️ Core Architecture

    Built on Anima: https://huggingface.co/circlestone-labs/Anima

    Uses the anima-preview3-base.safetensors variant as the foundation for a more stylized checkpoint fine-tuning with my own anime details dataset.

    Requires VAE and Text Encoder as separate files, following Anima's modular design.

    Fine-tuning is still under evaluation, but the model already delivers a more detailed anime style with consistent fidelity.

📦 Integrated Package

    Text Encoder (Qwen3) → https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders

    Qwen-Image VAE → https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae

    Unlike previous versions that consolidated everything into a single file, this release keeps the components separate for greater flexibility and control in environments such as ComfyUI.


🚧 Notes

    Fine-tuning is still being tested, but this version already produces a flatter, more normal anime style compared to experimental builds.

Files with SHA256: 45bb9bbd6772726f33373a27c74d3fcc8e61e34dac0f5e79e67305ed313c0257