Rendering text in AI images is the canonical hard problem. SD 1.5 essentially only produces gibberish; Flux and Ideogram are dramatic improvements, though still imperfect.
Per-model text rendering
| Model | Short words | Long sentences |
|---|---|---|
| SD 1.5 / SDXL | ★ (gibberish) | × |
| MJ v6 | ★★★ | ★★ |
| Flux Dev | ★★★★★ | ★★★★ |
| Ideogram 2.0 | ★★★★★ | ★★★★★ |
| DALL-E 3 | ★★★★ | ★★★ |
Writing tips
- Wrap text in quotes: text "FRESH MARKET"
- Specify font style: bold sans-serif / hand-written script
- Specify position: top center / bottom right corner
- Non-Latin scripts: most models still struggle; plan to overlay in post
Overlay workflow
For commercial product shots and posters where text matters, the reliable approach is:
- Prompt with "blank label area" / "space for text at top"
- AI generates the base image (no text)
- Add text in Photoshop / Figma / Canva
Industry practice: Professional commercial work nearly always uses AI for the base and a designer for the text — not direct AI text rendering.