OpenAIImage modelThe one that can spell.
Headlines, packaging copy, UI labels, signage. The words come out spelled correctly, in the font weight you asked for, including in scripts other models mangle.
Thumbnails and posters
Anything where the words are half the design. This is the model that gets the words right.
Packaging and labels
Product copy, ingredient panels, brand marks. Legible at the size a customer sees them.
Ad creative with a headline
The offer has to be readable in the feed. Generate the whole frame rather than compositing text on afterwards.
App and site mockups
Interface concepts with real labels, for a deck that has to look like a product and not a placeholder.
Localised creative
Non Latin scripts render properly, so one design ships in more than one market.
- Provider
- OpenAI
- Released
- April 2026
- Architecture
- Autoregressive, reasons before generating
- Resolution
- Up to 4K on Plus and above, longest edge 3840px
- Text rendering
- Best in class, including non Latin scripts
- Reference images
- Up to 4 per generation
- Transparency
- Transparent PNG on Plus and above
- Modes
- Text to image, image to image, editing
- Plans
- Every plan, including free
- Cost
- 3 to 9 credits per image
GPT Image 2.5 Flare
Almost always. Same credits, half the latency, better references. This is the default now.
GPT Image 2.5 Sunburst
For the shot that matters.
What is GPT Image 2?
GPT Image 2 is OpenAI's image generation model, released in April 2026. It is autoregressive rather than diffusion based, reasons about the prompt before generating, and is the strongest model available for rendering readable text inside an image.
Should I use GPT Image 2 or GPT Image 2.5 Flare?
Flare, in almost every case. It is the newer generation, it costs the same 3 to 9 credits, and it is faster and better at references. GPT Image 2 stays available for reproducibility: if a pipeline was tuned against it, older API ids still resolve to it.
How much does GPT Image 2 cost on APImage?
3 to 9 credits per image, depending on size and quality. It is included on every plan, including the free tier.
Can it render languages other than English?
Yes. Non Latin scripts were a long standing weak spot across the category and GPT Image 2 handles them, which is what makes it usable for localised creative.
Does it support transparent backgrounds?
Yes, transparent PNG output is available on Plus and above. For cutting out an existing image, the dedicated background removal tool is a flat 2 credits.
Can I edit an image I already have?
Yes. GPT Image 2 does image to image and editing, and takes up to four reference images per generation.
How do I use it?
Open the image studio and pick GPT Image 2. Describe the image, and put the exact words you want rendered in quotes. Add up to four reference images for style, product or likeness. Set the size, generate, and iterate on the prompt rather than in an editor.