mangadict models

ONNX models used by the mangadict browser extension to find and read Japanese text in manga pages. They are repackaged from the models below for in-browser inference with onnxruntime-web; see those repos for training details and their licenses (all Apache-2.0).

Path	Source	Notes
`detector/v4-s_int8.onnx`	ogkalu/comic-text-and-bubble-detector (`detector-v4-s_int8.onnx`)	Unmodified copy. RT-DETR; classes: bubble, text_bubble, text_free.
`ocr-2025-kv/encoder.onnx`, `cross.onnx`, `step.onnx`, `config.json`	jzhang533/manga-ocr-base-2025, a fine-tune of kha-white/manga-ocr-base	fp32 re-export with a KV cache: `encoder` (image → features), `cross` (per-image cross-attention keys/values), `step` (one decoder step: previous token + cache → logits + updated cache). Greedy decoding matches the original model.
`ocr-2025/vocab.txt`	kha-white/manga-ocr-base	Character vocabulary (shared by base and 2025).

Use tagged revisions (e.g. v1); main may change.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support