Generate image
Generate a still from a prompt and references, and pick the image model that fits the job.
You generate a new image in the generate bar. Light can set up the same shot for you if you'd rather describe it in plain words. See Ask Light.
Generate an image
Open + at the top of the left toolbar and choose Generate Image. An empty image shot appears on the board and the generate bar opens in the side panel.
- Pick a model. The model menu lists every image model, and on a free plan the premium ones show a lock.
- Write the prompt. Say what's in the shot, where it is and how it's lit, like "a white sneaker on wet concrete at night, low angle, neon reflections".
- Set the frame. Choose the aspect ratio, like 1:1 for a product page or 9:16 for a story, and the resolution, like 1K, 2K or 4K where the model offers it.
- Add references if you have them. A photo of your product keeps the shot true to it. Each slot has Upload for a file from your computer and Canvas to pick an image already on the board.
- Check the cost on the run button, which shows the credits next to an arrow, and press it.
The shot shows Generating… while it works, then the image takes its place. To try again, run it again and keep both, or use Edit and rerun to change one thing.
Some models have extra controls. GPT Image 2.5 has a quality setting that goes up to Max, Nano Banana 2.1 and Krea 3 have a thinking level for prompts with many parts, and the settings button holds less common options like a transparent background, web search and a negative prompt. The bar only shows what your model supports. See The generate bar for every control.
Pick an image model
Any image model can make a still from a prompt. They differ in look, size and how many references they take.
- For an edit that leans on several reference photos, use Nano Banana 2.1, GPT Image 2.5 or Krea 3.
- For a still at 4K, use Nano Banana Pro, Nano Banana 2 or Flux 3 Image.
- For a transparent background straight from the prompt, use GPT Image 2.5 Flare or Sunburst.
- For several options from one prompt, use Midjourney V7 or V8.1, which return a four-image grid.
- For vector art, use a Recraft row named Vector.
The full list is in Models.
What it costs
Every run spends credits, and the price on the run button changes with the model, resolution and quality. A run that fails gives its credits back. See Credits and When a run fails.