Some ideas start as a sentence. Others start as a photo you want to restyle or extend. The free Nano Banana tool on CanvaPro covers both paths: Text to Image and Image to Image. You do not need an account, and the usual download does not pick up a watermark from the page.
In either mode you write a prompt, choose an aspect ratio, complete the security check, and click Generate. A preview typically appears in about a minute so you can judge the result before you download. Image to Image is the path that also accepts a reference file: JPG, PNG, GIF, or WEBP up to 10 MB. GIF is unique to this uploader among the image generators in this batch, so it is worth using if your reference is already an animated or still GIF.
Text to Image versus Image to Image
Text to Image invents the frame from your description. There is no photo underneath. Image to Image starts from a file you upload and uses the prompt to guide the edit or restyle. If you already have a composition you like, Image to Image is usually the more controlled path. If you are starting from a blank idea, stay in Text to Image.
Aspect ratios on the page
The form offers Auto, 1:1, 3:2, 2:3, 9:16, 16:9, 3:4, and 4:3. Auto lets the model pick a fit. Use 1:1 for a square post, 9:16 for a story, 16:9 for a wide hero, and the 3:2, 2:3, 4:3, or 3:4 options when you want a more editorial still. Set the ratio before you generate so you are not cropping a finished image to fit the channel.
How to generate or edit an image, step by step
- Choose Text to Image or Image to Image. In Image to Image, upload a JPG, PNG, GIF, or WEBP that is 10 MB or smaller, and only use a file you have the right to edit.
- Write a prompt that names the subject, the setting, the light, and the style. One clear scene works better than a list of conflicting ideas.
- Pick an aspect ratio, complete the security check, and click Generate once.
- Review the preview. If the composition feels right, download it. If not, change one part of the prompt — or swap the reference — and run again. Keep earlier downloads so you can compare.
Choose text or image input, set the aspect, and when you retry, go back to the Nano Banana generator with a tighter prompt rather than a previous download.
Prompts that hold together
Name the object or person, the place, the time of day, and the mood. “A ceramic mug on a sunlit cafe table, shallow depth of field, warm film colors” is more useful than “nice photo.” Avoid packing tiny readable logos or dense paragraphs of type into the prompt if you need those letters to be accurate; generated typography is often the weak point. For Image to Image, say what should change and what should stay, so the model is not restyling the entire frame by accident.
Creators use Nano Banana for concept stills, product mock mood boards, and social drafts. Designers use Image to Image to explore a variation of a layout they already have. Treat every result as synthetic media, and check your own rights and labeling needs before a generated picture goes into an ad or a client deck.
A similar text-or-image flow, with different size options, is in the article on how to generate or edit images with Seedream.
If the preview looks off
Simplify the prompt, confirm the 10 MB limit in Image to Image, and make sure the security check is still valid. A busy queue can add a short wait. If you specifically want a GPT Image model instead of Nano Banana, switch to the GPT image generator, which has its own Text to Image and Image to Image modes and a separate set of model choices.
Uploads and prompts travel over HTTPS. Requests are handled to complete generation and are not sold for advertising profiles. Only upload reference photos you have the right to use.
A similar brief can be tried on the Seedream image generator if you want another model’s reading of the same idea.
After the Nano Banana preview appears
Judge the image at the size and aspect you will actually use. A wide 16:9 still can look generous on a desktop and still lose the subject when it is later cropped for a square post, which is why the ratio should be chosen on the form. If the composition is close, change one part of the prompt or adjust the reference and generate a new still, rather than heavily editing a near-miss in a way that fights the original frame.
Keep a short note of the mode, prompt, and aspect ratio next to the download. Treat the picture as synthetic media, and only use a reference photo you have the right to edit. When a client or platform needs a label, say that the image was generated or edited with AI.
You can also try the same brief in the walkthrough to generate images with GPT Image and keep the still that best matches the idea.
A short checklist
- Pick Text to Image, or Image to Image with a JPG, PNG, GIF, or WEBP under 10 MB.
- Write one clear prompt and choose Auto or a specific aspect ratio.
- Click Generate once and wait about a minute for the preview.
- Download the still you like, and keep earlier versions for comparison.
When you are ready, choose a mode, write a single concrete scene, and save the preview only after the composition looks right. One considered still is easier to use than a long string of slightly different experiments.