Text to image takes a written description and returns a still image. No source picture is involved: the prompt is the entire input, and the model composes subject, style, lighting and framing from scratch. It is the quickest way to get artwork that does not exist yet, which is why it turns up in product concepts, article headers, storyboard frames, ad variations, app icons and reference art handed to a designer.

Prompt wording does most of the work. Naming the subject, the medium, the lighting and the aspect ratio usually moves the output more than piling on adjectives, and generating a small batch then picking a winner beats refining one prompt forever. Text rendered inside an image remains the weak spot across the whole field, so treat any words in the output as a bonus rather than a guarantee and overlay critical copy yourself. Models also differ on resolution options, and some price the larger sizes differently.

On E2X this category currently carries Google's Nano Banana family, including the Pro and Lite variants, alongside OpenAI's GPT Image 2. Each is called the same way: send a model slug and an input object to the jobs endpoint, then poll the job id you get back until the result is ready. Swapping between them is a one-string change, and you pay per request from a prepaid balance, so comparing two models on the same prompt costs exactly two requests.

Frequently asked questions

Do I need a source image for text to image?

No. The prompt is the only required input and the model builds the picture from nothing. If you already have an image you want changed, the image to image category is the right one.

How am I charged for a generation?

Per request, drawn from your prepaid balance. Each model shows its own per-request price in the catalog before you call it, and some charge more for larger output sizes.

Can these models put readable text inside an image?

Sometimes, but no image model guarantees it. For headlines, prices or logos, generate the background and add the wording in a design tool afterwards.

What if I want to try a different model?

Change the model slug in your request. The endpoint, the authentication and the job polling flow stay identical across every model in the catalog.