Making a picture instead of a video
Switch Create to Image. GPT Image 2.5 by default, about 1 credit a picture, billed at what it costs.
At the top of Create, the Video / Image switch turns the composer into a picture maker, in every layout. Describe the picture, pick a model and a shape, and press Generate. The picture lands in your library under Images, and downloads as a PNG.
- GPT Image 2.5: about 1 credits a picture (the default). OpenAI's newest, and the best all-rounder in our tests: film and period looks, legible signs, and a face kept from one picture to the next. The default.
- GPT Image 2.5 HQ: about 4 credits a picture. The same model at high quality: finer detail and texture, about four times the price and half again the time.
- Seedream 4.5: about 4 credits a picture. ByteDance's, from the same family as the video model. Clean and photographic; the default until 16 October.
- Nano Banana Pro: about 15 credits a picture. Google's Gemini 3 Pro Image. Strongest at keeping a face and following a long description.
- Qwen-Image 2512: about 6 credits a picture. Alibaba's open model. The cheapest by far, and the best at lettering and signs; draws from words only, so head-and-shoulders and costume pictures are drawn with the default model.
Prices are estimates. You pay what the picture actually costs us, and never more than the estimate: it is charged when you press Generate, and the difference comes back to your balance as soon as the picture is ready.
Reference pictures: add up to 4 of your own pictures with Reference, or people from the cast and your characters with Cast, and the model draws from them and your words (each one adds a little to the estimate). A picture you upload can't show a person's face; for people, use the cast or your characters. Qwen-Image 2512 draws from words only.
Upscale to 4K makes the finished picture four times the size, for up to 1 credits more: a 16:9 picture comes back at 3840×2160. If the upscale fails, you get the picture as it was drawn and the upscale's credits back.
To try again with a change, open the picture and press Use this prompt again: Create opens in Image mode with the same prompt, model, shape and 4K choice, ready to edit. Reference pictures are not carried over; add them again.
Up to 4 pictures from one press, each charged on its own. On Free, pictures are drawn with GPT Image 2.5 and are not upscaled; the other models and the upscale come with a paid plan.
Pictures have to be of invented people. A prompt that names or asks for the likeness of someone real is refused before anything is charged.