feat: anime-text model

feat: better text rendering

chore(deps): update tokenizers to 0.23

fix: Revert inpainting logic for AOT/Lama (#609)

fix: ZLUDA compatibility issues (#582)

perf: prefer bf16

perf: FLUX.2 improvements

fix: only apply speech bubble mask logic to flux.2

perf: use small-decoder for flux-2-klein-4b

feat: Flux.2 Klein inpainting

chore: rust 1.95.0 clippy fix

perf: improve inpainting

fix: inpainting OOM & pipeline error handling

refactor(ml): shared TextRegion/TextDirection types for pipeline engines

refactor: remove bubbleDetector

feat: remove segmentation mask detections outside of detected text blocks (#378)

feat: add AMD GPU support via ZLUDA (#362)

tests: remove outdated

perf: improve ctd

feat: blob API, bubble/AOT-inpainting models, and rendering redesign

feat: custom tracing layer and log suppression

feat: pluggable engine system with DAG pipeline

refactor: simplify Tauri setup and dev server

feat: bootstrap window

refactor: merge crates

fix: pipeline crash when no text blocks detected (#291)

fix: shared llama.cpp backend

feat: llama.cpp-based paddleocr-vl-1.5

feat: integrate llama.cpp

feat: add open folder import, CUDA 13.2 build fixes, local LLM settin… (#271)

perf: fix paddleocr-vl

fix: align paddleocr-vl impl with transformers

chore: vendor paddleocr-vl

fix: remove normalize_ocr_text

fix: refine mask

fix: use f32 for paddleocr-vl

perf: speedup detection & ocr

feat: wire new models

feat: manga-text-segmentation-2025 model

feat: paddleocr-vl-1.5 model

feat: pp-doclayout-v3 model

feat: add bg-BG lang

feat: support more languages

fix: text box ordering regression

fix: compare_blocks_for_reading_order panics

refactor: simplify string serialization

feat: OpenAi Compatible

feat: add Deepseek provider (#235)

refactor: switch ws to http

ci: add cargo fmt