Note: Experimental version in testing. Fine tuning is in testing. A more normal, flat anime style is generated, but it is still being bug evaluated.
Changelog — FWAI ANIMA Magic
Core Architecture
Based on Anima https://huggingface.co/circlestone-labs/Anima, fine-tuned for a more detailed and official anime style.
Maintains the Anima encoding, but everything is bundled into a single .safetensors file.
This packaging includes both the Qwen CLIP and the Qwen-Image VAE, making the model heavier but easier to use across platforms.
Integrated Package
Files included in the package:
qwen_3_06b_base.safetensors
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders
qwen_image_vae.safetensors
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae
By consolidating everything into one file, users avoid handling multiple separate files, simplifying portability and integration in ComfyUI and other platforms.
Style and Tags
Integration of Danbooru-style tags, similar to Illustrious, for richer prompt control.
Improved recognition of character and pose tags, offering strong support for detailed anime outputs.
Enhanced fidelity in anatomical details and expressions, with more natural and consistent poses.
Usability and Workflow
The model follows a single-file workflow scheme, inspired by SDXL’s packaging idea, but not compatible with SDXL architecture.
Easier accessibility for users seeking a quick and reproducible setup.
🛠️ Technical Improvements
Optimized inference for prompts with multiple tags.
Reduced artifacts in outputs with complex backgrounds.
Better balance between sharpness and smoothness in anime-style renders.
This version makes it clear: the model is not SDXL-compatible, but adopts the single-file packaging idea to simplify distribution, including the VAE and CLIP inside one .safetensors file.
Description
Note: Experimental version in testing. Fine tuning is in testing. A more normal, flat anime style is generated, but it is still being bug evaluated.
Changelog — FWAI_ANIMA_Magic_v1.4B_(R5)
Core Architecture
Based on Anima https://huggingface.co/circlestone-labs/Anima, fine-tuned for a more detailed and official anime style.
Maintains the Anima encoding, but everything is bundled into a single .safetensors file.
This packaging includes both the Qwen CLIP and the Qwen-Image VAE, making the model heavier but easier to use across platforms.
Integrated Package
Files included in the package:
qwen_3_06b_base.safetensors
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/text_encoders
qwen_image_vae.safetensors
https://huggingface.co/circlestone-labs/Anima/tree/main/split_files/vae
By consolidating everything into one file, users avoid handling multiple separate files, simplifying portability and integration in ComfyUI and other platforms.
Style and Tags
Integration of Danbooru-style tags, similar to Illustrious, for richer prompt control.
Improved recognition of character and pose tags, offering strong support for detailed anime outputs.
Enhanced fidelity in anatomical details and expressions, with more natural and consistent poses.
Usability and Workflow
The model follows a single-file workflow scheme, inspired by SDXL’s packaging idea, but not compatible with SDXL architecture.
Easier accessibility for users seeking a quick and reproducible setup.
🛠️ Technical Improvements
Optimized inference for prompts with multiple tags.
Reduced artifacts in outputs with complex backgrounds.
Better balance between sharpness and smoothness in anime-style renders.
This version makes it clear: the model is not SDXL-compatible, but adopts the single-file packaging idea to simplify distribution, including the VAE and CLIP inside one .safetensors file.