gpt-image-edit

To edit embedded multilingual text while preserving layout, use identity-preserving image editing via GPT Image 2. You provide the exact target characters in your prompt, and the model rewrites in-image headlines or labels while keeping typography and framing stable.

Yes, you can use multiple reference images for controlled composition. Provide up to 10 input image URLs with clear per-image intent to blend subject identity, lighting, and scene elements into the final edited output.

The best way to localize a headline and CTA while keeping the brand mark is applying a targeted image-to-image edit. This approach rewrites text across scripts like Latin, CJK, or Arabic while keeping the original subject and framing stable.

Yes, GPT Image 2 editing supports non-Latin scripts including CJK, Cyrillic, and Arabic. You can rewrite in-image text by quoting the exact characters needed, ensuring the localized text matches the original layout typography.

When sending an edit request, the size parameter is optional and constrained to auto or supported fixed presets. You must include a prompt and an images array containing your reference URLs to satisfy execution requirements.

Choose identity-preserving image editing when you need to modify existing visuals with precise text or layout changes while keeping the subject stable. Route to text-to-image generation when you need to create entirely new visuals or perform batch generation instead.