GPT Image 2.5 Flare
OpenAI's fast default; text, layouts and edits, up to 16 references. Your reference photos reach it. The credits per still show in the composer's picker before you send.
GPT Image 2.5
One photo in, six formats out
One bottle. Six ads. The label spelled every time.
Drawn onGPT Image 2.5 Flare, edit, high
Text in the frame
Three pieces of type, spelled from one prompt.
One text-to-image prompt on Sunburst, at high quality, asked for a can labelled 'VOLT' in a burst of ice, a skier mid-flip behind it, the headline 'CHARGE THE DROP' in heavy condensed type and 'VOLT ENERGY' at the foot, and no other text. All three came back as written, and nothing else was lettered.
- Quote the words and say where they sit.
- Read them back before the still goes anywhere; a model can still slip.
GPT Image 2.5 in the Studio
Up to 16 reference photos
Drop the product, the room, the person into the composer; they go to GPT Image 2.5's edit endpoint with the prompt, up to sixteen at once.
Words spelled as written
Quote the label, the headline, the small print. The colourway labels and the poster's three lines on this page came back exactly as asked.
Edits that keep the frame
Send a still back with one change and the camera, the light and every object stay put; only what you named moves.
Flare or Sunburst
Flare is OpenAI's fast default; Sunburst is the precision tier for fine detail and a face or product kept across edits. Both are in the picker.
Ten frames, exact pixels
1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9 or 21:9, sent to GPT Image 2.5 as an exact width and height.
Credits before you send
Drawn at high quality and priced per still; the picker shows the credits for each variant before you send.
Cutouts, from the model
GPT Image 2.5 can return a PNG with a transparent background, as the cutout above shows. The studio does not ask for one yet.
Every format on the wall
Each still lands on your Assets wall with its prompt, to compare, star, sort into folders and download.
Make it an element, then shoot it
Make element turns the ad's product into a reference a scene is held to, so the bottle in the poster is the bottle in the clip.
FAQs about GPT Image 2.5
8 Image Models
GPT Image 2.5 Sunburst
OpenAI's precision tier; fine detail and identity kept across edits, up to 16 references. Your reference photos reach it. The credits per still show in the composer's picker before you send.
GPT Image 2
OpenAI's image model; strong on text and instructions, takes up to 16 references. Your reference photos reach it. The credits per still show in the composer's picker before you send.
8 Image Models
Every image model the composer draws on, Nano Banana Pro, FLUX.2 Pro, Seedream 4.5, Ideogram 4.5, FLUX 1.1 Pro included, each priced per still in the picker.
How to make a set of ads
1. Attach the product
Sign in, switch the composer to Image and drop in the product photo. Add a room, a model or a texture if the ad needs them.
2. One line per format
Pick GPT Image 2.5 Flare or Sunburst and a frame, then ask for one format at a time: the scene, the colourways, the poster with its words in quotes.
3. Keep the set
Every still lands on your Assets wall. Star the keepers, download them, or press Make element and put the product in a scene.
Your first set
One photo in. A set of ads out.
Attach the product, ask for one format at a time, read the words back.