ByteDance

Seedream 4.5 creates images from text and edits them from as many as 14 reference images. ByteDance emphasizes faithful preservation of faces, lighting, color, and other reference details, along with improved multi-image composition and readable dense text. The Omnipix integration returns one 2K or 4K image per request and supports preset aspect ratios or the proportions of the first reference.

Seedream 5 accepts a text prompt for new images and optional image references for editing or style guidance. Pro configurations expose 1K and 2K output, while Lite is oriented around 2K and 3K output and includes separate generation and editing entries in this catalog. Up to ten references may be used, but fewer well-chosen images usually make each reference's purpose easier to communicate.

Seedance 1.5 Pro is ByteDance's joint audio-video model for text-driven and image-driven clips. It is designed for prompts that combine action, camera direction, dialogue, ambient sound, and music, and can use first and last frames to constrain the visual transition.

Seedance 2.5 supports video generation from text, a first frame, first and last frames, or collections of image, video, and audio references. It produces clips with optional synchronized sound and explicit duration controls. Seedance 2.0 remains available in Full, Fast, and Mini variants for alternative quality, speed, and cost profiles.