Poster_editor
Edit a poster without the original file — remove text and elements and rewrite them with the same font, colour and shadow, even by plain-language request.
The problem
A poster almost always arrives as a flat image: changing a date, a name or a typo needs the designer’s original file, which isn’t there. Retouching it by hand means erasing the text without leaving smudges and rewriting it identically, font, outline and shadow included.
The solution
- It detects text blocks by itself (OCR) and measures their style, colour, outline and shadow.
- It erases an area in three ways — solid fill, patch, AI reconstruction — and suggests the right one.
- Magic wand and lasso to select, copy and move objects.
- It recognises the font from the image, or honestly says none is convincing.
- It can be driven in plain language: “move the date under the title and make it yellow”.
- It fixes a punctuation mark without rewriting the line.
How it’s built
- A single rendering engine, Konva.js, for both preview and export: the preview is the result, it doesn’t just resemble it.
- The Python backend never draws text: it only touches the source pixels (erasing, OCR, selection).
- Text layout is chosen by measuring the fill: Python proposes candidates, the browser decides with the font’s real metrics.
- The language model never sees the image: it reads a sheet of measured facts, so it can’t make things up.
Screenshots
