bytedance/seedream-5-pro

OfficialView on Replicate →

ByteDance's flagship text-to-image and image editing model, generating sharp 1K and 2K images from text or up to 10 reference images

Capabilities

Reference Images

Cost

Community model (estimated from hardware time)

Input Parameters

Name	Type	Description	Default	Constraints
`aspect_ratio`	string	Image aspect ratio (ignored in layer decomposition mode). Use 'match_input_image' to automatically match the input image's aspect ratio.	`"match_input_image"`	match_input_image1:14:33:416:99:163:22:321:9
`image_input`	array	Input image(s). In standard mode, 1-10 reference images for image-to-image generation. In layer decomposition mode, exactly one image to decompose.		—
`layer_decomposition`	boolean	Split a single input image into a base image plus separate element layers (up to 16). Requires exactly one input image. Each returned layer includes its stacking order, bounding box, name, and description.	`false`	—
`output_format`	string	Output image format. In layer decomposition mode this only controls the base image; layers are always returned as PNG.	`"png"`	pngjpeg
`prompt`	string	Text prompt. In standard mode, describes the image to generate. In layer decomposition mode it is optional: describe which elements to split out, or leave blank to auto-detect all major elements. Maximum 4000 characters.	`""`	—
`size`	string	Image resolution. Standard mode supports 1K and 2K. Layer decomposition mode supports 1K, 1.5K, 2K, and auto; auto sizes the base image and layers from the input image.	`"2K"`	1K1.5K2Kauto

aspect_ratiostring

Image aspect ratio (ignored in layer decomposition mode). Use 'match_input_image' to automatically match the input image's aspect ratio.

Default: "match_input_image"

match_input_image1:14:33:416:99:163:22:321:9

image_inputarray

Input image(s). In standard mode, 1-10 reference images for image-to-image generation. In layer decomposition mode, exactly one image to decompose.

Default:

layer_decompositionboolean

Split a single input image into a base image plus separate element layers (up to 16). Requires exactly one input image. Each returned layer includes its stacking order, bounding box, name, and description.

Default: false

output_formatstring

Output image format. In layer decomposition mode this only controls the base image; layers are always returned as PNG.

Default: "png"

pngjpeg

promptstring

Text prompt. In standard mode, describes the image to generate. In layer decomposition mode it is optional: describe which elements to split out, or leave blank to auto-detect all major elements. Maximum 4000 characters.

Default: ""

sizestring

Image resolution. Standard mode supports 1K and 2K. Layer decomposition mode supports 1K, 1.5K, 2K, and auto; auto sizes the base image and layers from the input image.

Default: "2K"

1K1.5K2Kauto

Version: 91daad99b90fUpdated: 9/20/2026183.9K runs