Literal instruction following
Reviewers note it gets the details other models skip, like a glass filled to the brim or a clock at an exact time. Long prompts with many parts mostly land.
GPT Image · by OpenAI
GPT Image generator
GPT Image is OpenAI's image model, the one behind ChatGPT's image generation. GPT Image 2.5 is known for following long, specific instructions to the letter, putting the right words in speech bubbles and labels, and editing with many reference images. Describe an image below, or add a photo to edit, and it opens in Image Studio on GPT Image 2.5 Flare.
Example image
Opens in Image Studio with the model you picked. You see the price in credits before anything runs.
What GPT Image is known for
Words in the right place
Speech bubbles, labels and poster lines come out spelled as written. Our four-panel comic got both lines of dialogue in the right bubbles.
Edits with many references
Give it up to 16 reference images, or mark the area to change, and describe the edit. Everything else stays put.
Two flavours: Flare and Sunburst
Flare is the fast default; Sunburst is slower and aimed at precise edits and fine detail. They cost the same.
Where it's weaker
- It can repeat a pose across panels: our four-step push-up guide got every label right but showed the same plank in all four.
- Complex prompts can take up to two minutes, OpenAI says.
- Recurring characters and brand elements can drift between separate images.
- Exact layout (where each element sits) is weaker than its text accuracy.
- Images of real people in sensitive or deceptive contexts are refused; outputs carry C2PA metadata and an invisible watermark.
- Maker
- OpenAI · GPT Image 2.5 September 2026
- Versions on Affogato
- 2.5 Flare · 2.5 Sunburst · GPT Image 2
- Modes
- Text to image · edit with up to 16 references or a mask
- Resolution
- Up to 3840 px on the long edge
- Transparent backgrounds
- Yes (PNG, WebP)
- Aspect ratios
- 9:16 · 3:4 · 1:1 · 4:3 · 16:9
Made with GPT Image on Affogato
Every image below was generated with the prompt shown, unedited. Load any prompt into the generator and make your own version.
Comics with dialogueA four-panel cafe comic
Prompt
A four-panel comic strip, stacked vertically. Panel 1: a sleepy customer in pajamas shuffles into a cafe. Panel 2: the barista asks, speech bubble: "The usual?" Panel 3: the customer, eyes half closed, answers: "The biggest one." Panel 4: the barista slides over a comically huge mug. Clean line art, soft colors.
Creators on cameraA vlogger on a rainy Tokyo street
Prompt
Candid vertical photo of a 24-year-old creator with a short wolf-cut filming a vlog on a rainy Tokyo street at night, holding a small camera at arm's length and grinning into it. Neon reflections on wet pavement, natural skin texture, shot on a phone.
Same face, new sceneThe Tokyo vlogger, moved on stage
Prompt
Keep the exact same person from the first image: same face, haircut and smile. Now they are on stage at a tech conference, holding a microphone and presenting, a large screen glowing behind them.
Labels right, poses repeatFour labelled panels, one pose
Prompt
A clean fitness infographic: a female coach demonstrates a push-up in four labeled panels, "1. Plank", "2. Lower", "3. Press", "4. Breathe". Each panel shows the same woman in the same black outfit from the side. White background, thin dividers, simple sans-serif labels, no other text.
Poster with a personA jazz night poster
Prompt
A concert poster featuring a jazz saxophonist in a sharp burgundy suit mid-solo under a single spotlight. The headline at the top reads "BLUE HOUR SESSIONS". Below the image: "Friday 12 December · Doors 8 PM · The Velvet Room". Bold modern typography, film-grain texture.
UGC with a labelA protein shaker selfie
Prompt
UGC-style vertical selfie of a man in his twenties at a gym holding a matte black protein shaker up to the camera. The bottle reads "FORGE" and, below it, "30g PROTEIN" in bold white letters. Natural sweat, overhead gym lighting, slightly crooked phone framing.
Fashion on a personCobalt trench in Paris
Prompt
Full-body street-style photo of a woman in her fifties with silver hair in a tailored cobalt-blue trench coat, wide cream trousers and loafers, crossing a sunlit Paris boulevard mid-stride. Framed like a fashion-week street photographer's shot.
Group photosFive friends on a rooftop
Prompt
Five friends in their twenties laughing on a rooftop at sunset, one pouring lemonade from a glass jug, string lights overhead. Candid documentary style, every face distinct and natural.
Which GPT Image to pick
Start here
GPT Image 2.5 Flare
Most images
The fast default. OpenAI says it beats GPT Image 2 with up to half the wait.
11 credits per image
Open with GPT Image 2.5 FlareOption 2
GPT Image 2.5 Sunburst
Precise edits and fine detail
Slower than Flare, and tuned for edits where small things have to be exactly right.
11 credits per image
Open with GPT Image 2.5 SunburstOption 3
OpenAI GPT Image 2
Matching earlier work
April 2026. OpenAI retires it on 10 February 2027, so move new work to 2.5.
43 credits per image
Open with OpenAI GPT Image 2OpenAI released GPT Image 2.6 on 4 November 2026; it is not on Affogato yet, and this page will say when it arrives. Credits are read from the live catalog; the Standard plan ($24/month) includes 2,000 credits, about 181 GPT Image 2.5 Flare images.
Prompting GPT Image well
- Organise long prompts into parts: the scene, the subject, the details, and anything that must not appear.
- Put exact text in quotes and say its position and style; add "no other text" to keep the rest clean.
- For a photo look, say "real photograph" and name the light, lens and materials.
- For edits, state the one change, then list what must stay: the face, the pose, the background.
- When panels must differ, describe each panel's pose or action separately rather than relying on the label.
From prompt to image
Describe the image or add a photo
Write what you want in the box at the top of this page, or start from an example. Add a photo to edit it instead.
Sign in and land in Image Studio
Your prompt, photo, format and the GPT Image version you picked carry over. Nothing runs until you press Generate.
Check the price, then generate
The credit price shows under the Generate button before anything is spent.
Refine with an edit
Send the result to Edit and change one thing at a time; switch to Sunburst for fine, exact edits.
Questions about GPT Image
What is GPT Image best at?
Following long, specific instructions, putting the right words into comics, labels and posters, and edits that use many reference images.
What is the difference between GPT Image 2.5 Flare and Sunburst?
They are two versions of the same model. Flare is faster and the everyday default; Sunburst is slower and aimed at precise edits and fine detail.
Is GPT Image the same as ChatGPT's image generator?
Yes. GPT Image is the model behind image generation in ChatGPT; on Affogato you use it directly, alongside other image and video models on one credit balance.
Is GPT Image 2.6 available?
OpenAI released it on 4 November 2026. It is not on Affogato yet; GPT Image 2.5 Flare and Sunburst are.
Is GPT Image free on Affogato?
No. Every image costs credits, shown before you generate. Credits come with a monthly plan and unused credits roll over up to the plan's cap.
Can I use GPT Image images commercially?
Yes. Paid Affogato plans include a commercial license, within our terms of service and OpenAI's usage policies.
Keep exploring
Facts checked November 15, 2026. Sources: fal: GPT Image 2.5 · OpenAI: image generation guide · OpenAI: image prompting guide · OpenAI: ChatGPT Images 2.5 system card · OpenAI: deprecations · MindStudio: GPT Image 2.5 review