ImageTextTools

• 

  Detected text regions are numbered on the image; type new wording on the right and the app redraws them in place.

• 

  Each edit returns three versions; compare the details and download the one that reads best.

What it does

ImageTextTools replaces the text inside a photo while keeping the original type, layout and background. Upload an image, the text is detected automatically, you type the new words, and you get 3 redrawn versions to pick from. Best for signage, labels, price tags, screenshots and short marketing lines.

The problem

Retyping text inside an image usually means rebuilding it in Photoshop — the existing letters can't simply be retyped, and OCR + font-matching tools break on stylized type. ImageTextTools repaints the whole text region conditioned on the surrounding pixels, so it never needs to know the font.

How we built it

• TanStack Start (React 19) on Cloudflare Workers, with D1 + R2
• Google Cloud Vision for text region detection
• Generative image models (Grok, Nano Banana, Qwen via Kie.ai) for redrawing the text area
• SeeAPI moderation on every input and output
• Fully serverless: one active edit per visitor, idempotent credit handling, a cron reconciler that rescues stuck jobs

Challenges

Matching the surrounding typography without knowing the font was the core problem — redrawing the whole region instead of compositing letters turned out to be the answer. Building a crash-safe generation pipeline on serverless (retries, stuck-job rescue, webhook idempotency) was the second one.

Try it

Every account gets one free HD download, no card required.

Built With

• cloudflare-workers
• google-cloud-vision
• react
• typescript