Feature lists are easy to inflate. What matters is whether a feature removes a real step from your workflow: writing the prompt, controlling the frame, checking the text, or finding the result after you leave the page.
This page covers the capabilities that are actually exposed in the generator, the settings that change the output, and the limits worth knowing before you spend credits.
Core generation modes
Four entry points, each with a different input and credit cost.
Text to image
Describe a scene and generate it from scratch. Costs 3 credits at 1K, 6 at 2K, and 12 at 4K.
Image to image
Upload a reference image and describe the change. Costs 6, 12, and 24 credits at the same resolutions.
Old photo restoration
A dedicated workflow for scratches, fading, and colorization using one uploaded photo. One free 1K restoration is available after email verification.
Birthday doodle maker
Add handwritten-style birthday wishes and face-safe doodles to a portrait, with title, date, and message controls.
Text rendering and layout control
Short, structured text is where the model performs best.
GPT Image 2 can render readable typography inside an image, which makes it useful for posters, labels, infographics, and social graphics. The result depends heavily on how constrained the text is. A headline with a clear hierarchy is a different task from a paragraph of body copy.
- Put exact copy in quotation marks and state the language.
- Name the position: top center, inside a label, under the product, or across a banner.
- Describe the hierarchy: headline, subhead, and small supporting labels.
- Keep supporting copy short. Long paragraphs and tiny legal text still need line-by-line checking.
- Inspect spelling and character shapes after every generation, especially for non-Latin scripts.
For infographics, specify the reading order and the number of sections. A defined grid produces a more predictable layout than an open-ended request for a poster.
Resolution and aspect ratio
Two settings that are often confused, with very different effects on price and composition.
Resolution
- Options
- 1K, 2K, 4K
- What changes
- Level of detail and credit cost; 1K is about 1024px, and higher resolutions cost more.
Aspect ratio
- Options
- 1:1, 16:9, 9:16, 4:3, 3:4
- What changes
- Frame shape and composition; the credit cost stays the same.
Mode
- Options
- Text to image, image to image
- What changes
- Whether a reference image is used; image-to-image costs double.
| Setting | Options | What changes |
|---|---|---|
| Resolution | 1K, 2K, 4K | Level of detail and credit cost; 1K is about 1024px, and higher resolutions cost more. |
| Aspect ratio | 1:1, 16:9, 9:16, 4:3, 3:4 | Frame shape and composition; the credit cost stays the same. |
| Mode | Text to image, image to image | Whether a reference image is used; image-to-image costs double. |
Choose the aspect ratio before writing the prompt. A square social post and a 16:9 web hero need different compositions, even when the subject is the same.
A prompt library that removes the blank page
Start from a complete prompt instead of guessing at structure.
The prompt library collects complete GPT Image 2 prompts with the settings used to create each example. Detail pages include why the prompt works, which parts are meant to be replaced, and practical tips for adapting it.
- Copy the full prompt or open it directly in the generator.
- See the mode, aspect ratio, and resolution used for each example.
- Use the notes to adapt variables without breaking the structure.
- Move between related prompts when you need a different style for the same subject.
The library is a starting point, not a gallery of guaranteed results. Output still varies with the subject, text, and resolution you choose.
Long-running tasks and task history
Image generation can outlast a page visit, so the task does not depend on the tab staying open.
Submitting an image creates a task that continues after you leave the generator. The task center shows pending, processing, and completed tasks, along with the prompt and settings used. You can return later to retrieve a result instead of regenerating and paying twice.
- Close the page or switch devices without cancelling the task.
- Check that a result is complete before starting another generation.
- Keep the prompt and settings attached to each task for later reference.
- Use the task history to compare iterations without losing earlier outputs.
Where GPT Image 2 still needs a human check
The honest limits matter more than another bullet point about creativity.
- Long strings of text, small type, and unusual character sets can contain errors.
- Hands, small faces, and repeated objects sometimes need another generation.
- Image-to-image is a generative edit. It is not a pixel-perfect retouch or object removal tool.
- Brand-accurate logos, packaging copy, and regulated claims need manual verification.
- Dense multi-column layouts with small body copy are harder to control than a single clear hierarchy.
- Results vary by prompt. Treat the first output as a draft and budget for one refinement.
These limits are not a reason to avoid the tool. They are the parts of the workflow that deserve a review step before the image goes into production.
Pricing is visible before you generate
No surprise deduction after the image arrives.
The generator shows the credit cost for the selected mode and resolution before a task is submitted. Paid credits are deducted after the upstream task is accepted, and a failed task is refunded to the original balance.
Plan prices, effective cost per credit, and per-image estimates are documented separately on the pricing page and in the cost guide. That keeps the product features and the billing rules in two places, each with a single source of truth.
GPT Image 2 feature FAQ
Quick answers about resolutions, editing, text, and task handling.
Does GPT Image 2 support 4K output?+
Yes. Text-to-image supports 1K, 2K, and 4K at 3, 6, and 12 credits. Image-to-image supports the same resolutions at 6, 12, and 24 credits.
Can I edit an uploaded photo?+
Yes. Image-to-image mode uses an uploaded image as a reference and applies a described change. It is a generative edit, not a pixel-level retouch, so review details you did not ask to change.
Does text render reliably?+
Short text with a clear hierarchy renders more reliably than long paragraphs. Put exact copy in quotation marks, state the language and position, and inspect the result before using it.
Is there a prompt library?+
Yes. The library includes complete prompts with the mode, aspect ratio, resolution, notes on why each prompt works, and tips for adapting the variables.
Can I leave the page while an image is generating?+
Yes. The task continues after you leave. Open the task center to check progress or retrieve the completed image.
Is this the official OpenAI product?+
No. This is an independent GPT Image 2 generator and is not affiliated with or endorsed by OpenAI.
Keep reading
See the features in use
Generate a free 1K image
Verify your email and test text rendering, composition, and the task flow with one free 1K text-to-image image.
This is an independent GPT Image 2 generator and is not affiliated with or endorsed by OpenAI. Feature availability, credit costs, and output quality can vary by mode and resolution.
Last updated: October 2026