OpenAI · Image generation and editing
GPT-Image-2 — precise prompt adherence and masked edits
The model to reach for when the brief is exact and the text has to be readable.
Try GPT-Image-2What is GPT-Image-2?
GPT-Image-2 is OpenAI’s current image model, available on Floniks for both generation and editing. Its distinguishing trait is literal prompt adherence — it does what the brief says rather than what looks good in general, which matters when a marketing visual has to match a spec. The editing mode accepts an optional mask so you can confine a change to one region, and text-to-image supports transparent backgrounds for logos and icons.
Available on Floniks
Credit cost per render, read live from the model catalogue. All models draw from one balance — no per-model subscription.
- OpenAI GPT-Image210 credits
- OpenAI GPT-Image-2 (Text to Image)10 credits
What GPT-Image-2 is best for
- Marketing visuals that must match an exact written brief
- Logos and icons that need a transparent background
- Local retouching where a mask confines the edit to one region
- Product image cleanup
How to use GPT-Image-2
Choose generate or edit
Text-to-image starts from a prompt; editing takes up to 10 input images plus an optional mask.
Add a mask for local edits
White marks the area to repaint. Without a mask the model may touch the whole frame.
Be literal in the prompt
This model rewards specificity. Vague direction produces generic output.
Frequently asked questions
- Does GPT-Image-2 support transparent backgrounds?
- Yes, in text-to-image mode — useful for logos, icons and overlays that need to sit on arbitrary backgrounds.
- How do masked edits work?
- Supply the original image plus a mask where white marks the region to repaint. Everything outside the mask is preserved.
- GPT-Image-2 or Nano Banana for editing?
- GPT-Image-2 when the instruction is precise and text legibility matters. Nano Banana when the edit is creative and open-ended.
