Seedream is a family of image generation models from the ByteDance Seed team. Seedream 5.0 Lite, the version listed here, is a unified multimodal model: one neural network that takes text and images as input and produces images, so the same model creates new pictures and edits existing ones.
The model follows long, detailed instructions on style and layout, for example for posters, slides, app screens and charts. It has an online search capability, so a prompt can ask for current news, weather or prices and the image shows them. It also infers the intent behind a prompt; one example on the maker's page asks it to annotate the calories of the food in a photo. Sequential generation produces a series of related images from one request.
Developers call the model through ModelArk, the model API platform of BytePlus, ByteDance's enterprise cloud. Use is billed per output image, and input images are free.