What is GPT Image 2?
GPT Image 2 is OpenAI's image model. It makes new pictures from a text prompt and edits the photos you give it. OpenAI introduced it on April 21, 2026, together with ChatGPT Images 2.0, and offers it to developers in its API as gpt-image-2.
OpenAI built it for detailed briefs: objects placed where you ask, dense text that stays readable, and designed layouts such as posters, explainers, diagrams and comics. It reports significant gains in text written in non-Latin scripts, especially Japanese, Korean, Chinese, Hindi and Bengali, and gives the model a knowledge cutoff of December 2025.
On Bolly it runs in your browser on the Free plan, with no ChatGPT subscription and no OpenAI API key. Bolly is a separate service and isn't affiliated with OpenAI.
GPT Image 2 free: what the daily credits cover
Sign in with Google and your account gets 10 free credits a day. The quality setting decides what each image costs:
- low: 1 credit an image, ten free images a day
- medium: 3 credits an image, three free images a day
- high, the default: 8 credits an image, one free image a day
Auto lets the model pick the quality and is priced like high. Size, background and file format don't change the price, and a batch costs the same as making its images one at a time.
Unused credits don't roll over: the balance goes back to 10 at midnight UTC. Everything you make on the Free plan is public and may be featured in Explore. Private images need a paid plan. Starter is $19 a month with 300 credits, which covers 37 images a month at high quality. See plans.
Your first character swap video is free too. It unlocks 24 hours after sign-up, runs on Viggle, has a Bolly watermark and uses 5 of that day's credits.
How to use GPT Image 2 online
Write the brief
Describe the image in the box above: subject, setting, style and light. Put any words it should show in quotes and say where they go.
Add photos to edit
Attach one or more images with the + button, or drop them on the box, and the composer switches to Edit. Say what to change and what to leave alone.
Pick quality and count
Choose auto, low, medium or high quality and 1 to 4 images. Image size, background and file format are in the settings next to the credit price.
Generate and download
Press the arrow and sign in with Google if asked. Images appear under the box with Download and Reuse prompt buttons, and in your Library.
Prompts that play to its strengths
OpenAI built GPT Image 2 to follow detailed instructions, so write the prompt like a brief: the exact words in quotes, where each element sits, and what must not appear. Two to try:
A clean educational poster titled "How Honey Is Made" in bold dark-brown letters across the top. Below the title, five numbered steps run down the page, each with a small flat illustration and a one-line caption: "Bees gather nectar", "Enzymes start the change", "Wings fan it dry", "Honey fills the comb", "Wax caps seal it". Warm yellow and cream palette, plenty of white space, no other text.
A summer night-market poster with the Japanese headline "夏祭り" in large brush lettering at the top, and the English line "Summer Night Market" below it in a clean sans-serif. Paper lanterns, food stalls and a crowd in light summer kimonos under a deep blue evening sky. Spell all text exactly as written and add no other words.
Edit photos with GPT Image 2
Attach a photo and describe one change at a time, then list what has to stay as it is: the face, the product, the label, the light. With more than one image, refer to each by its order and its role, such as the product in the first image and the room in the second. One edit takes up to 16 images.
More photos raise the price a little. At high quality a one-photo edit is 9 credits, and a day's free credits cover an edit with up to five photos. At low quality, a one-photo edit is 2 credits.
Keep the bottle in this photo exactly as it is, including its shape, label and color. Replace the background with smooth pale sand in soft morning sun, add a short shadow to the right of the bottle, and write "Made for long days" in white sans-serif in the top left corner. Change nothing else.
New images start at landscape 4:3, and an edit keeps whatever size is set. To keep your photo's own shape, set Image size to auto before you run the edit.
Quality, sizes and transparent backgrounds
- Quality: auto, low, medium or high, with high as the default. Use low for drafts and high for the final image.
- Images per run: 1 to 4, all from the same prompt.
- Image size: square HD, square, portrait 4:3, portrait 16:9, landscape 4:3, landscape 16:9 or auto. Every size costs the same.
- Background: auto, transparent or opaque. Transparent gives a cutout with no backdrop, for logos, stickers and product shots; keep the file format on PNG or WebP for it. OpenAI still lists transparency on GPT Image 2 as a preview.
- File format: jpeg, png or webp, with png as the default.
What it can still get wrong
OpenAI lists the gaps itself. The model can struggle with tasks that need a coherent physical world, like origami guides or a Rubik's Cube, with details on hidden, angled or reversed surfaces, and with very dense, repeating texture such as grains of sand. Labels and arrows in diagrams may need a check. Read every word before you publish, and fix a wrong detail with a follow-up edit.
GPT Image 2 and ChatGPT
If you came looking for the ChatGPT image generator: ChatGPT is OpenAI's app, and Bolly runs OpenAI's API model in its own composer, with its own credits and Free plan. In ChatGPT, a thinking model can also search the web and double-check its images. Here you prompt the image model directly and see the price before each run.
ChatGPT has offered Images 2.5 since September 8, 2026, and both of OpenAI's API models from that launch are on Bolly: GPT Image 2.5, the Sunburst model, and GPT Image 2.5 Flare. Bolly doesn't run the older GPT Image 1.5, which OpenAI plans to shut down on December 1, 2026.
An app built on GPT Image 2
Web Designer runs GPT Image 2 with a website-mockup prompt already written. Describe a site and it draws a landing page inside a browser frame, at high quality for 8 credits.
Photos of people
Only edit photos of yourself, or of people who said yes, and never of anyone under 18. When an image looks real, say it's AI-made where you post it. Bolly's Responsible AI policy has the details.
Frequently asked questions
Is GPT Image 2 free?
Yes, on Bolly's Free plan. Every account gets 10 credits a day, and a GPT Image 2 image costs 8 at the default high quality, 3 at medium and 1 at low. You don't add a card, and the credits come back every day. Paid plans add monthly credits and private images.
How many free images do I get a day?
With 10 credits: one image at high quality, three at medium or ten at low. You can mix them: after one high-quality image, the credits left cover two low-quality drafts.
Are my free images private?
No. On the Free plan every image is public and can be featured in Explore. Paid plans let you keep generations private and add folders.
How much does GPT Image 2 cost on Bolly?
8 credits an image at high quality, 3 at medium and 1 at low. Edits add a little for the photos you attach: 9 credits with one photo at high. The credits come from the Free plan's 10 a day or from a paid plan; Starter is $19 a month with 300 credits.
What sizes and quality settings does GPT Image 2 offer?
Quality is auto, low, medium or high, starting on high. Image size offers square HD, square, portrait 4:3, portrait 16:9, landscape 4:3, landscape 16:9 or auto, with landscape 4:3 for new images. Each run makes 1 to 4 images, saved as jpeg, png or webp.
Can GPT Image 2 edit my photos?
Yes. Attach one or more photos, up to 16, and the composer switches to Edit. Say what to change and what to keep. A one-photo edit costs 9 credits at high quality and 2 at low.
Can GPT Image 2 make transparent backgrounds?
Yes. In the settings next to the credit price, set Background to transparent and keep the format on PNG or WebP. OpenAI still calls this a preview on GPT Image 2, so check fine edges such as hair and glass. A transparent image costs the same as any other.
GPT Image 2 vs GPT Image 2.5: which should I use?
GPT Image 2.5 is OpenAI's newer model. On Bolly it runs Sunburst, which OpenAI tuned for precise edits and fine detail, and it adds xhigh and max quality. At low, medium or high the two cost the same here. Stay on GPT Image 2 when its results already work for you, or try GPT Image 2.5 Flare when speed matters.
GPT Image 2 vs Nano Banana 2: which should I use?
Both are on the Free plan. Nano Banana 2, Google's model, is 4 credits at its default 1K, so it covers two images a day, and it goes up to 4K with an optional web search. GPT Image 2 adds quality levels and a transparent background setting. For text-heavy designs, run the same prompt on both and keep the one you like.
Is this the ChatGPT image generator?
No. ChatGPT is OpenAI's app, and Bolly isn't affiliated with OpenAI. GPT Image 2 is OpenAI's image model, which OpenAI also offers in its API. Bolly runs that model with its own credits and Free plan, so you don't need a ChatGPT account to use it.






