When you want a still from a written scene, or a variation of a photo you already have, the free GPT image generator on CanvaPro gives you Text to Image and Image to Image in the same form. You pick a GPT Image model, write a prompt, optionally add a reference, and generate a preview. There is no account to create, and the usual download does not pick up a watermark from the page.
The live model list is GPT Image 2, GPT Image 2.5 Sunburst, and GPT Image 2.5 Flare. Aspect ratios are Auto, 1:1, 9:16, 16:9, 4:3, and 3:4. If you use Image to Image, the reference can be a JPG, PNG, or WEBP, with a live size limit of 10 MB. After you complete the security check and click Generate Image, a preview appears. Jobs are queued and often take a few minutes, depending on load.
What you can and cannot pick on the page
You can choose the model and the aspect ratio. You cannot pick a 1K, 2K, or 4K output size on the live form. There is no visible resolution control, so do not expect a separate 2K or 4K switch. Work with the aspect ratio that matches where the image will live, then download the preview if you like the composition.
Text to Image and Image to Image
Text to Image invents the frame from the prompt. Image to Image starts from a reference photo you upload and uses the prompt to guide the result. Use a reference only when you have the right to edit that file. A clear still with a simple composition usually steers the model more reliably than a busy, tiny thumbnail.
How to generate an image, step by step
- Choose Text to Image or Image to Image, then pick GPT Image 2, Sunburst, or Flare. If you are comparing models, change only the model between runs so you can see the difference clearly.
- Write a prompt that names the subject, setting, light, and style. In Image to Image, also say what should change and what should stay, and upload a JPG, PNG, or WEBP under 10 MB.
- Pick an aspect ratio, complete the security check, and click Generate Image once. Leave the tab open while the job sits in the queue.
- Review the preview and download it if the composition works. Keep earlier files so you can compare models or prompt tweaks side by side.
Choose text or image input, write a clear prompt, and download the preview only after you have checked it; a retry on the GPT Image page should start from that improved sentence.
Prompts and practical uses
Specific stills hold up better than vague requests. Name materials, time of day, and camera distance if they matter. Avoid relying on tiny generated typography if the letters have to be exact. Creators use the tool for concept art, product mood images, and social drafts. Teams use it to mock a scene before a shoot. Treat every output as synthetic media, and check your own rights and labeling needs before a picture is used in advertising or a client deliverable.
If you are starting from a picture you already have, it often helps to extract a prompt from the image and then edit that sentence before you generate.
If the job is slow or the still looks wrong
Confirm the security check is still valid, and in Image to Image confirm the 10 MB limit and file type. A busy queue can add a few minutes, which is expected on this page. If you would rather try a different generator with a broader aspect list and GIF support on the reference, open the Nano Banana tool and run the same prompt there for a second opinion.
Uploads and prompts travel over HTTPS. Requests are handled to complete generation and are not sold. Only upload reference photos you have the right to use.
A sentence taken from a reference still with the image to prompt tool can give you a tighter brief before you generate again.
After the GPT Image preview appears
Open the still at the size it will be published. Check the main subject, the edges of the frame, and any typography you hoped would be readable. If letters are wrong, it is usually better to add real type in a design app than to keep prompting for a perfect sign. If the composition is almost right, change one variable — the model, the aspect ratio, or a single phrase in the prompt — and run a new job from the same starting point.
Save the file with a name that includes the model you used, and keep earlier previews for comparison. Jobs on this page are often queued for a few minutes, so one careful prompt is a better use of that wait than several nearly identical submissions. Treat the result as synthetic media, and only upload a reference you have the right to edit.
A different text-to-image form, with its own aspect choices, is covered in how to generate images with Grok Imagine.
A short checklist
- Pick GPT Image 2, Sunburst, or Flare, plus Auto or a specific aspect ratio.
- Write a clear prompt, and optionally add a JPG, PNG, or WEBP under 10 MB.
- Click Generate Image once and wait for the queued preview, often a few minutes.
- Download the still you like. There is no separate 2K or 4K picker on the form.
When you are ready, choose one model, write a single concrete scene, and save the preview only after it looks right. A finished still is easier to reuse than several overlapping jobs from the same prompt.