The short answer
- GPT Image 2 is OpenAI's image model for generating and editing product ads from text and reference photos. It was snapshotted as gpt-image-2-2026-04-21 and has since been followed by GPT Image 2.5 and 2.6.
- Per OpenAI's guide on November 9, 2026, a 1024x1024 image costs $0.006 (low), $0.053 (medium) or $0.211 (high). Landscape and portrait images cost $0.005, $0.041 and $0.165.
- It supports custom sizes up to 3840 pixels on an edge (4K), so one model covers 1:1, 4:5, 9:16 and 16:9 placements.
- OpenAI says text placement and brand consistency across generations can still fail. Review every output.
What GPT Image 2 is, and what came after it
OpenAI's model page calls GPT Image 2 its state-of-the-art image generation model for fast, high-quality generation and editing. It takes text and images in and returns images. You reach it through the Image API (generations and edits) or as a tool inside the Responses API, which supports multi-turn editing.
It is no longer the newest model. OpenAI's image generation guide now leads with GPT Image 2.5, in two versions, and lists GPT Image 2.6.
| Model | Aimed at | Notes from OpenAI's guide |
|---|---|---|
| gpt-image-2 | General generation and edits | Quality levels low, medium, high. Always processes input images at high fidelity. |
| gpt-image-2.5-flare | Fast everyday generation | Adds xhigh and max quality levels. $30 per million image output tokens. |
| gpt-image-2.5-sunburst | Workflows where editing precision matters most | Same token rates as Flare. Adds xhigh and max quality. |
| GPT Image 2.6 (Sunburst and Flare) | Detailed work and fast iteration | Adds a complexity control, markup-guided edits, SVG output and animated GIF output. |
The practical read: use the model name in your tool or API call only after checking which generation it is. A tool that says "GPT Image 2" may be on the older snapshot, which is slower and does not offer the newer quality levels.
GPT Image 2 pricing per image
As listed in OpenAI's image generation guide on November 9, 2026. Output is billed in image tokens. These are output costs per image; text and image input tokens are billed on top.
| Quality | 1024x1024 | 1024x1536 portrait | 1536x1024 landscape |
|---|---|---|---|
| Low | $0.006 | $0.005 | $0.005 |
| Medium | $0.053 | $0.041 | $0.041 |
| High | $0.211 | $0.165 | $0.165 |
Worked example: 100 landscape ad variants cost about $0.50 at low quality, $4.10 at medium and $16.50 at high, before input tokens. Use low for layout drafts, medium for variant testing, high for the final.
Edits with reference photos cost more on the input side. OpenAI notes that because the model always processes input images at high fidelity, image input tokens can be higher for edit requests that include reference images.
Sizes that match ad placements
OpenAI's constraints for gpt-image-2: both edges must be multiples of 16, the longest edge can be at most 3840 pixels, the ratio of long to short edge cannot exceed 3:1, and the total pixels must fall between 655,360 and 8,294,400. Square images are typically the fastest to generate.
| Placement | Ratio | A size that fits the rules |
|---|---|---|
| Feed square | 1:1 | 1024x1024 |
| Feed portrait | 4:5 | 1024x1280 |
| Stories, Reels, TikTok | 9:16 | 1152x2048 |
| Landscape, YouTube, display | 16:9 | 2048x1152 (OpenAI's 2K landscape) |
| Wide banner | 3:1 | 1536x512 (exactly 3:1, 786,432 pixels) |
Standard social sizes such as 1080x1350 are not multiples of 16, so generate at a nearby valid size and let the ad platform scale it.
How to use it for product ads
- Start from the real product photo. Send it as an input image to the edits endpoint. Describe the scene, not the product.
- Draft at low quality. Try six scenes for pennies. Pick two.
- Write the text separately. Put exact words in quotes in the prompt: a four-word headline and a short badge are far safer than a paragraph.
- Re-render the winners at high quality at the placement sizes in the table above.
- Review against the product. Check the label, color, proportions and every character of text.
A sample prompt for a running shoe: "Edit this photo: place the shoe on wet asphalt at dawn, low angle, soft rim light, keep the shoe, laces and logo exactly as in the input, headline 'Run lighter.' in white at the top left." The instruction to keep the shoe unchanged is what protects the product.
What OpenAI says it still gets wrong
- Latency. Complex prompts can take up to 2 minutes.
- Text rendering. Significantly improved, but it can still struggle with precise text placement and clarity.
- Consistency. It may struggle to keep a recurring character or brand element visually consistent across generations.
- Composition control. Structured, layout-sensitive compositions can be hard to place exactly.
Those are OpenAI's own limitations notes. They are why ad teams generate many options and review each, rather than trusting one render.
How it compares
Google's Nano Banana family is the main alternative. It offers a lower price per image on the Flash models and up to 14 reference images; see Nano Banana 2 for ad creatives. For targeted edits of product photos, see Flux Kontext for product photos. To turn the winning still into motion, see best image to video AI generators for ads.
Where Sprites fits
Sprites makes image ads and ad copy in chat. Its image ads are rendered with OpenAI's GPT Image 2.5 models: Flare for drafts and Sunburst for finals. Every ad is held for your approval before it goes live in Google or Meta. See the AI static ad generator.
Build is $99 a month, with a 7-day trial for $1. See pricing.
FAQ
What is GPT Image 2?
GPT Image 2 is OpenAI's image generation and editing model, available in the API as gpt-image-2 with the snapshot gpt-image-2-2026-04-21. OpenAI describes it as a state-of-the-art model for fast, high-quality generation and editing that accepts text and image input. It outputs images only.
How much does GPT Image 2 cost per image?
Per OpenAI's image generation guide on November 9, 2026, a 1024x1024 image costs $0.006 at low quality, $0.053 at medium and $0.211 at high. Portrait and landscape images at 1024x1536 or 1536x1024 cost $0.005, $0.041 and $0.165. Input text and image tokens are billed on top.
What image sizes does GPT Image 2 support?
GPT Image 2 accepts any size where both edges are multiples of 16, the longest edge is at most 3840 pixels, the long-to-short ratio is at most 3:1, and the total pixels are between 655,360 and 8,294,400. That covers 1:1, 4:5, 9:16 and 16:9 ad sizes, up to 4K.
Is GPT Image 2 good at text in ads?
It is much better than earlier models, but not perfect. OpenAI's own limitations list says the model can still struggle with precise text placement and clarity, and with keeping recurring brand elements consistent across generations. Check every word on every image before it runs.
Is there a newer model than GPT Image 2?
Yes. OpenAI's image generation guide now centers on GPT Image 2.5 (gpt-image-2.5-sunburst and gpt-image-2.5-flare) and lists GPT Image 2.6. Sunburst is aimed at editing precision and Flare at fast everyday generation. The guide tells new integrations to use the newer models.
Can GPT Image 2 make product ads with my real product?
Yes, by editing. You send your product photo as an input image and describe the scene, and the model processes input images at high fidelity automatically. Always compare the label, color and shape against the real product before using the result.