Getting started
What makes a good source image
OCR quality is the single biggest determinant of translation quality. Resolution, contrast and image type all matter — here's what to aim for.
Resolution
72 dpi at 1024 × 1024 or higher is the sweet spot for most use cases. Higher resolution gives the OCR clearer character shapes; very low resolution can collapse similar glyphs (Chinese 千/干, Latin l/I etc.).
Contrast and background
High contrast between text and background is ideal. Busy gradient backgrounds (typical of game UI or marketing banners) work but may require manual touch-up in the editor.
Image type and what to expect
- Screenshots: usually near-perfect — clean glyphs, predictable layouts.
- Manga / comic panels: good with Manga Mode, which handles speech bubbles and vertical text.
- Product photos: solid in E-commerce Mode; expect minor tweaks for stylised typography.
- Photos of street signs / menus: variable — strong lighting and a frontal angle help a lot.
- Scanned documents: best at ≥200 dpi scan resolution; multi-column layouts are reconstructed automatically.
If you have the original asset (Figma, PSD, vector source), exporting at 2x resolution and re-uploading usually beats trying to upscale a screenshot.