mayocream/anime-text-yolo

Anime Text YOLO

YOLO12 model weights for detecting text blocks in anime, manga, and illustrated images. Five model-size variants are provided as safe, framework-neutral PyTorch state dictionaries.

Model details

• Task: object detection
• Architecture: Ultralytics YOLO12
• Class: text_block
• Format: SafeTensors state dictionary
• Intended loader: a compatible YOLO12 implementation using the architecture metadata stored in each file

Files

File	Variant	Tensors	Size
yolo12n_animetext.safetensors	YOLO12n	691	9.95 MiB
yolo12s_animetext.safetensors	YOLO12s	691	35.54 MiB
yolo12m_animetext.safetensors	YOLO12m	751	77.14 MiB
yolo12l_animetext.safetensors	YOLO12l	1,245	101.17 MiB
yolo12x_animetext.safetensors	YOLO12x	1,245	226.22 MiB

Each file header contains the class names, YOLO architecture YAML, training arguments, conversion date, source path, and Ultralytics version.

Loading

These files are raw state dictionaries, not pickled Ultralytics checkpoint objects. Instantiate the matching YOLO12 architecture before loading a file.

from safetensors.torch import load_file

state_dict = load_file("yolo12n_animetext.safetensors")

Intended use and limitations

Use these weights to locate rectangular text regions before OCR or translation. The training dataset and evaluation results are not included, so validate recall, precision, and class behavior on the target image distribution before production use. The files are not directly loadable with YOLO("file.safetensors") without reconstructing the model architecture.

License

GPL-3.0. See LICENSE.