OpenAI · the ChatGPT Images 2.0 model
GPT Image 2 is the model OpenAI launched as ChatGPT Images 2.0. Turn a detailed brief into posters, infographics, comic pages, and app mockups with readable text, or upload up to 16 images to edit and combine. No ChatGPT plan or API key needed, and drafts start at 2 credits.
Text-heavy layouts
GPT Image 2 is at its best when an image has to carry words: headlines, labels, speech bubbles, buttons. Each example lists the ratio, size, and quality we would start with.
Suggested: 3:4 · 2K · medium · 16 credits
Put every line that must appear in quotes, from the headline and a second-language title to the date and venue, and GPT Image 2 sets them into the layout as written, as on this kite-festival poster. Before printing, check each line against your prompt and make sure no extra words crept in.
Suggested: 16:9 · 2K · medium · 32 credits
Upload a photo of notes or a whiteboard as Image 1 and ask for the same steps as a finished slide: the text carries over while the layout, icons, and colors are redesigned. Check that every label, number, and unit survived the redraw; the corner card shows the source sketch.
Suggested: 3:4 · 2K · medium · 16 credits
Describe each panel and quote every speech bubble and sound effect, and the model lays out the whole page in one image. Check that the characters match from panel to panel, each bubble reads as written, and a spot color stays on the object you gave it to.
Suggested: 9:16 · 1K · medium · 16 credits
List the greeting, cards, buttons, stats, and tab labels in quotes, and you get a believable screen for a pitch or a quick flow test. Treat it as a mockup: icons and spacing are approximate, so rebuild the screen in a design tool before it ships.
These habits come from OpenAI's prompting guide and common failure reports. They matter most when the image has to carry words.
Quote the exact words
Put every word that must appear in quotes, say where it goes, and add “no other text”. Spell unusual names letter by letter.
Write the brief in order
Scene, then subject, then key details, then constraints. Name the use, such as “event poster”, “infographic slide”, or “app screen”, so the layout follows.
Draft at low, finish higher
Check layout and spelling at low quality, then rerun the same prompt at medium or high for small text, dense panels, or faces.
Label every reference
Write “Image 1: whiteboard photo. Image 2: color palette.”, then “change only X; keep Y the same”, and repeat the keep-list each round.
Paragraphs of small print
Long copy squeezed into a corner is where typos appear. Split text across panels or keep it to a few lines, and proofread before you publish.
Exact counts or pixel placement
“Exactly 10 slices” or “logo 12 px from the edge” may not hold; OpenAI lists precise placement as a known limit.
Transparency or masks
This workbench has no transparent-background or mask setting. Generate on a plain backdrop, then use Background Remover.
Recent events or live data
There's no web search here, and OpenAI gives the model a December 2025 knowledge cutoff. Write the facts you need into the prompt.
Fix one wrong word with an edit: upload the image as Image 1 and write “Change only the date line to SAT 19 APRIL · 10:00–17:00. Keep the layout, colors, kite, and every other word exactly the same.”
OpenAI released GPT Image 2.5 Flare and Sunburst on September 8, 2026. Credits follow output cost, so quality names don't line up: GPT Image 2 medium costs the same as GPT Image 2.5 high (16 credits per 1K image), and GPT Image 2 high matches 2.5 max (60).
| Compare | GPT Image 2 | GPT Image 2.5 Flare | GPT Image 2.5 Sunburst | Nano Banana Pro |
|---|---|---|---|---|
| Released | April 21, 2026 | September 8, 2026 | September 8, 2026 | November 20, 2025 |
| Quality levels | low · medium · high | low · medium · high · xhigh · max | low · medium · high · xhigh · max | One level |
| Credits per 1K image, by level | 2 · 16 · 60 | 2 · 4 · 16 · 32 · 60 | 2 · 4 · 16 · 32 · 60 | 30 |
| Reference images, max | 16 | 16 | 16 | 14 |
| Choose it for | Text layouts, or a 2.0-era series | New images and everyday edits | Edits where the rest must not move | Five-person shots and brand assets |
OpenAI treats GPT Image 2 output larger than 2560×1440 as experimental, so 4K and the largest 2K sizes can vary more.
Each model has its own page and runs in the same workbench on one credit balance.
GoogleGoogle's previous Flash image model, kept for continuing an earlier series and reusing prompts tuned for it.
View model
OpenAIFaster everyday generation, natural lighting and rich textures
View model
OpenAIPrecision-first detail, premium visual fidelity
View model
NewGoogleConsistent characters, product placement, exact short text
View modelRemove the background, upscale, vectorize or denoise any result, all on the same credit balance.
Transparent PNG or WebP with clean hair and product edges, ready for a listing or a new composition.
2× / 4× / 8× with recovered texture instead of smeared pixels, so a 1K draft can become a print.
Logos, icons and flat illustrations become editable SVG that scales to any size.
Remove grain and compression artifacts from low-light photos and scans while keeping real detail.
Yes. GPT Image 2 (API name gpt-image-2) is the OpenAI image model released on April 21, 2026 and launched in ChatGPT as ChatGPT Images 2.0. It creates images from text and edits images you upload, with a focus on readable text, structured layouts, and non-Latin scripts. OpenAI has since released GPT Image 2.5 (Flare and Sunburst), which ChatGPT now uses; GPT Image 2 stays available on Oeneo AI as its own model.
For new work, start with GPT Image 2.5 Flare; OpenAI says it gives higher quality than GPT Image 2 at about half the latency. Use GPT Image 2.5 Sunburst when an edit must leave the rest of the image untouched. Choose GPT Image 2 when you're continuing a series made with ChatGPT Images 2.0, reusing prompts written for it, or want a second take on a text-heavy poster or infographic. The comparison table on this page lists credits for every quality level.
Yes. Oeneo AI runs GPT Image 2 for you, so you don't need a ChatGPT plan, an OpenAI account, an API key, or organization verification. Write a prompt on this page, add reference images if you need them, and generate. Your prompt is sent as written, without a chat assistant rewriting it first. Oeneo AI is a browser workbench and doesn't offer its own API.
You can try it free. New accounts get 10 credits, valid for 365 days, and a low-quality text-to-image result costs 2 credits at 1K or 2K, so the bonus covers about 5 images. Images made on the free plan are public; paid plans keep them private by default.
Text to image costs 2–3 credits per image at low, 16–24 at medium, and 60–90 at high, from 1K to 4K; editing costs a little more and is listed in the specs. Higher quality spends more rendering effort, which helps small text, dense panels, and faces, and it takes longer; OpenAI notes complex prompts can take up to two minutes. The Generate button shows the exact cost before you run, and failed tasks are refunded automatically.
Yes. The 4K setting goes up to 3840×2160 at 16:9 and 2880×2880 at 1:1. OpenAI treats GPT Image 2 output larger than 2560×1440 as experimental, so 4K results vary more. For print, generating at 2K and enlarging with Image Upscaler is the more predictable route.
Yes. OpenAI says GPT Image 2 improved non-Latin text such as Chinese, Japanese, Korean, Hindi, and Bengali. Short headlines and labels work best; small or dense characters can still come out wrong, so check every word. Put the exact text in quotes, and use medium or high when the text is small.
Yes. Upload up to 16 images, say what each one is (“Image 1: product photo, Image 2: style reference”), and describe the change. Write “change only…” and list what must stay the same. There's no mask brush, and each edit is a new request, so download a result and upload it again for the next round. When the rest of the image must stay untouched, GPT Image 2.5 Sunburst is built for that.
Not on Oeneo AI. OpenAI offers transparent output for GPT Image 2 only as an API preview, and this workbench doesn't expose that setting. Generate the subject on a plain background, then use Background Remover to get a transparent PNG.
Yes, on paid plans: subscriptions include a full commercial license, and the Lifetime plan covers personal commercial use (see Pricing). Oeneo AI adds no visible watermark; OpenAI says it uses C2PA metadata and invisible watermarking to help identify images made with its tools. Whether a specific image is safe to publish still depends on what it shows, such as real people, brands, or artwork.
Start with one poster
Quote the exact words, pick a quality, and generate. New accounts start with 10 free credits.
