WUMO comic translator
Scrapes a daily comic, translates the lettering with Claude vision and re-letters the image.
WUMO Collector & Translator fetches the featured comic strip from the WUMO front page, detects genuinely new strips via perceptual hashing, sends the image to Claude for OCR plus a...
WUMO Collector & Translator fetches the featured comic strip from the WUMO front page, detects genuinely new strips via perceptual hashing, sends the image to Claude for OCR plus a humour-preserving Russian translation with bounding boxes, and repaints the translation onto a copy of the image.
The overlay stage uses a colour-keyed glyph mask, connected-component filtering and background-anchored inpainting to erase original lettering without damaging artwork.
A FastAPI gallery shows original and translated strips side by side; a scheduler and Telegram notifier run in Docker.
What it does
- Scraper targeting only the featured strip; pHash + Hamming distance dedup
- Claude vision structured-JSON output: text blocks, boxes, funny RU translation
- Image re-lettering pipeline: ink-colour mask, border-touching component drop, inpaint, auto-fit wrapped Cyrillic text
- Paged web gallery with on-demand run trigger; console/Telegram notifications
- Background interval worker; persistent image volume; one-command release scripts
Practical multimodal LLM pipeline (vision OCR + translation + layout) combined with hand-built image processing for clean re-lettering. Note: sample images are a third-party copyrighted comic; use only as a small before/after illustration or blur.
Want something like this?
I build it live on a Zoom call, you watch the screen, the timer stops when you say stop. First 15 minutes are free.