How to create your first AI image
Six steps from an empty form to an image you are willing to publish. Every limit quoted here comes from the workspace's own configuration, not from a marketing page.
1. Decide what you are starting from
Before touching the form, separate two questions: what you are starting from, and what you want to end up with. Starting from an idea, from a photo you already have, and from reference material are three different runs inside the same workspace.
Getting this wrong is the most expensive mistake in the sequence, because it costs a whole run rather than a rerun.
- Name the element that must survive. A reference image without that sentence is a suggestion, not a constraint.
- The workspace states each model's reference-image limit inside the form. Read it there rather than assuming a number.
| What you have | What that makes it | What to attach |
|---|---|---|
| An idea and nothing else | Text to image | Nothing |
| A photo you want changed | Image editing | That photo, as a reference image |
| Something the result must match | Reference-guided generation | The reference images, and the element that has to survive |
2. Pick a model and let the picker set the limits
The picker starts on Nano Banana Pro. Which aspect ratios, resolution tiers and reference-image limits are on offer is decided by that choice, so change the model before you invest in the prompt rather than after.
The models differ in more than quality. A tier that is missing the aspect ratio you need is not a worse model for your job, it is the wrong one.
- Check the reference-image limit if your brief depends on attachments.
- The model pages each carry samples and published prompts; the picker is the authority on what your run will accept.
3. Write the brief in one pass
Write what is in the frame before writing how it should look. A prompt that names a subject, its position and the light gives the model something to be wrong about, which is what makes a second attempt productive instead of random.
Mention the destination. A brief written for a square product card and a brief written for a wide banner lead to different framing decisions.
- Name the subject and the one thing about it that matters.
- Say where it is and what is behind it.
- Say how it is lit.
- Say what the image is for.
4. Set the format and the resolution
The aspect ratio is a shape decision and the resolution tier is a detail budget. Neither repairs a vague brief, and neither adds something the prompt never asked for.
- 9:16 for a phone screen, 16:9 for a video thumbnail, 1:1 for a profile image.
- Pick the resolution the destination can actually use. A larger file is a larger file, not automatically a better picture.
- Do not use resolution to chase detail the brief did not specify. Go back to the brief instead.
5. Read the quote, then send
The form quotes the price of the run you are about to send, and it is set by the service rather than by the page. Resolution tier, reference images and the model all move it, so read the figure against what you are actually asking for.
6. Review it before you publish
Treat every result as a draft until it fits the brief. Look at the full size rather than the thumbnail, because the things that break are the things a thumbnail hides.
- Hands, faces and jewellery are where structure usually gives way.
- Check small type by reading it, not by glancing at it.
- Scan the reflections and the edges for anything the prompt did not ask for.
- Before publishing, confirm you may use both what you attached and what you got back.
Questions
Which model should a beginner pick?
Nano Banana Pro, which is what the picker starts on. Its resolution tiers and its reference-image limit cover most first briefs, and you can change model without leaving the form.
Can I change the model after writing the prompt?
Yes. The picker sits in the same form as the prompt, so switch it and send again. The quote is recalculated for whichever model you actually send.
The result ignored part of what I asked for. Why?
Usually because the brief described a look instead of a scene. Name what is in the frame first and how it should look second. A short prompt that says one specific thing tends to hold better than a long one that says five vague ones.
Why does this guide not list prices?
Because the cost of a run depends on the model, the resolution tier and the reference images you attach, and all three are things the form knows when you send it. Read the quote in the form rather than a figure on a page.
Can I use the result commercially?
This guide grants no rights. Whether you may use a result commercially depends on the terms of the model you ran, on the rights attached to anything you uploaded, and on this site's own terms. Check all three before you publish.
Run the six steps yourself
The workspace is on the image generator page. Take your own brief through it.
Open the image generator