GPT Image 2 rewards specific instructions. The difference between a usable image and three discarded ones is usually not a secret prompt trick; it is deciding the frame, the resolution, and the exact text before you submit.
This walkthrough follows the actual workflow in the generator, from email verification to downloading the final file. Each step also notes where credits are spent and where they are easy to waste.
Before your first generation
Three things to sort out first, so the first image is a real test instead of a throwaway.
- Create an account and verify your email. A new verified account receives enough credits for one free 1K text-to-image image, with no card required.
- Decide whether you need text-to-image or image-to-image. The second mode requires an uploaded reference image and costs twice as many credits.
- Pick the output resolution before you write the prompt. A 4K image costs four times as many credits as 1K, so save it for finals.
If you only want to test how the model handles a style, start from a prompt in the prompt library. Each example includes the full prompt, settings, and notes on what makes it work.
The seven-step workflow
Follow this order the first few times. Once the settings are familiar, steps 5 and 6 can be adjusted per image.
- 1
Open the generator
Start on the homepage generator or open the full workspace. The homepage version is enough for a first test; the workspace gives the generator more room.
- 2
Sign in and verify your email
Verification unlocks the free 1K text-to-image generation. Sign in before writing a long prompt so you do not lose it at submission.
- 3
Choose text-to-image or image-to-image
Use text-to-image when the image comes from a description. Use image-to-image when an existing image must stay in the result; upload the reference first.
- 4
Write the prompt
Describe the subject, composition, light, style, and any text that must appear. Put exact copy in quotation marks and keep it short.
- 5
Set resolution and aspect ratio
Choose 1K, 2K, or 4K, then pick 1:1, 16:9, 9:16, 4:3, or 3:4. Resolution changes the credit cost; aspect ratio changes the frame without changing the price.
- 6
Generate and review
Check the result against the prompt, not against a vague idea. Inspect hands, text, edges, and repeated objects before deciding what to change.
- 7
Refine or download
If the composition works, keep it and fix one issue at a time. Download the resolution that matches the final use instead of regenerating blindly.
Leaving the page does not cancel a running task. Open the task center to check progress or retrieve a result later.
What a useful prompt contains
A prompt is a set of constraints. Add the ones that prevent the most likely wrong interpretation.
Short prompts are not automatically better. A prompt needs enough information to lock the parts that matter: who or what is in the frame, where it sits, how it is lit, what style it should follow, and which text must be exact.
- Subject and action: what is in the image and what it is doing.
- Composition: close-up, full-body, centered, wide shot, or negative space for headlines.
- Lighting: soft window light, hard flash, golden hour, studio softbox, or flat overcast light.
- Style and material: editorial photograph, hand-drawn marker, watercolor, 3D render, or paper cutout.
- Exact text: the words in quotation marks, the language, and where they appear.
- Constraints: no logo, no extra props, no hands, plain background, or keep the original pose.
Example prompt
A single ceramic coffee bag on a walnut table, morning window light from the left, 50mm product photograph, muted beige background, label text "SINGLE ORIGIN" in clean uppercase sans-serif, no hands, no extra props.
Change one element at a time when you iterate. If you rewrite the whole prompt after every result, you cannot tell which instruction fixed the problem.
Getting text to render correctly
Text rendering is one of the main reasons people choose GPT Image 2. It still needs a constrained prompt.
- Keep the text short. One headline and one supporting line are more reliable than a paragraph.
- Put the exact words in quotation marks and state the language.
- Say where the text belongs: top center, inside a label, or under the product.
- Describe hierarchy explicitly: headline, subhead, and small label sizes.
- Avoid tiny legal copy and dense body text unless you plan to check every line.
- Inspect spelling after generation. If it fails, regenerate with a narrower instruction instead of adding more text.
For posters and infographics, specify the reading order. A clear top-to-bottom hierarchy is easier to render than a layout where every element competes for attention.
Editing an existing image
Image-to-image keeps a reference in the result, but it is still a generative edit, not a pixel-level retouch.
Upload one reference, then describe only the change you want. Repeat the parts that must stay: the face, pose, background, framing, or color palette. The model needs both sides of the instruction: what changes and what does not.
- Use image-to-image when the result must stay recognizably tied to the uploaded image.
- Keep the edit instruction short and specific. One change per generation is easier to evaluate.
- Expect 6 credits at 1K, 12 at 2K, and 24 at 4K, double the text-to-image cost.
- For scratches, fading, and colorization of old photos, use the dedicated old photo restoration tool instead of a general edit.
- Do not expect pixel-perfect removal or retouching. Generative edits can alter details you did not mention.
Resolution, aspect ratio, and credits
Resolution and mode set the credit cost. Aspect ratio changes the frame shape without changing the price.
1K
- Text-to-image credits
- 3
- Image-to-image credits
- 6
- Typical use
- Drafts, social previews, quick composition tests
2K
- Text-to-image credits
- 6
- Image-to-image credits
- 12
- Typical use
- Web hero images, sharper product shots, presentation visuals
4K
- Text-to-image credits
- 12
- Image-to-image credits
- 24
- Typical use
- Print, large displays, and images with dense detail
| Output | Text-to-image credits | Image-to-image credits | Typical use |
|---|---|---|---|
| 1K | 3 | 6 | Drafts, social previews, quick composition tests |
| 2K | 6 | 12 | Web hero images, sharper product shots, presentation visuals |
| 4K | 12 | 24 | Print, large displays, and images with dense detail |
Available aspect ratios are 1:1, 16:9, 9:16, 4:3, and 3:4. Use the cost guide for the full plan-by-plan price math.
Common mistakes that waste credits
Most wasted generations come from changing too much at once or choosing the wrong output size.
Rewriting the entire prompt after one bad result
Change the subject, composition, lighting, or text constraint separately. Otherwise you cannot tell what worked.
Asking for a paragraph of text
Short headlines and labels render more reliably than long body copy. Reserve long text for layouts you can inspect line by line.
Choosing 4K for a draft
4K costs four times as many credits as 1K. Use 1K to decide the composition, then generate the final at the resolution you need.
Using image-to-image to explore
Image-to-image costs double. Explore the concept in text-to-image first, then switch to editing when a specific reference must be preserved.
Ignoring the aspect ratio
A great image in the wrong frame still needs another generation. Decide whether the final use is square, landscape, portrait, or poster before submitting.
Treating the first result as final
Budget for one or two refinements. A second focused generation is usually cheaper than forcing a flawed image into the final asset.
GPT Image 2 usage FAQ
Short answers about the free image, modes, resolutions, and text rendering.
Is GPT Image 2 free to try?+
Yes. After verifying your email, a new account can generate one free 1K text-to-image image without a card. Image-to-image and higher resolutions use credits.
What is the difference between text-to-image and image-to-image?+
Text-to-image creates an image from a written description. Image-to-image uses an uploaded image as a reference and applies a described change. It costs twice as many credits: 6, 12, and 24 at 1K, 2K, and 4K.
Which resolution should I choose?+
Use 1K for drafts, social previews, and composition tests; 2K for web hero images and sharper product visuals; 4K for print, large displays, and dense detail. Higher resolutions cost more credits.
Can I close the page while an image is generating?+
Yes. The task continues after you leave the page. Open the task center to check progress or retrieve the result.
Why does my text look wrong?+
Long copy, unquoted text, and unclear placement are the usual causes. Keep the text short, put the exact words in quotation marks, state the language and position, then regenerate with a narrower instruction.
How many credits does one image cost?+
Text-to-image costs 3 credits at 1K, 6 at 2K, and 12 at 4K. Image-to-image costs 6, 12, and 24 at the same resolutions. The generator shows the exact cost before you submit.
Keep reading
Start with the free image
Write one prompt and generate it at 1K
Verify your email, open the generator, and test the workflow with the free 1K text-to-image image before buying credits.
This is an independent GPT Image 2 generator and is not affiliated with or endorsed by OpenAI. Results vary by prompt, mode, and resolution; review each output before using it commercially.
Last updated: October 2026