Seedream 4.5

Photorealistic frames, legible text inside the image, and wide aspect ratios.

modellane/seedream-4.5 · Text to image · ByteDance

An image model that turns text into high-resolution, photorealistic frames. It is strong at texture, skin, and material detail; at building natural light and depth; and at rendering legible lettering and typography inside the image. It holds a composition together across wide aspect ratios, which is why it is the pick for horizontal work such as cover images and banners.

Vendor

ByteDance

Task

Text to image

Output

Image

Price

$0.06 · per image

Input parameters

• Prompt prompt — required, at most 5000 characters Describe what you want produced as concretely as you can.
• Size size — optional, default: 2048*2048, options: 2048*2048, 2560*1440, 1440*2560, 2304*1728, 1728*2304

Recommended use cases

• Product and editorial photography — Studio-lit product frames for catalogs, landing pages, and decks; material and surface detail are preserved.
• Images with text and typography — Short headlines set legibly inside posters, covers, and social media images.
• Wide compositions — Full-frame output for wide and tall ratios: web hero images, banners, and video covers.
• Concept and pre-production — Compare location, set, and atmosphere options quickly at high resolution.

Where it is strong

• Realistic texture, light, and material detail
• Writes short text inside the image legibly
• Keeps the framing across wide and tall aspect ratios
• Delivers a single high-resolution frame

Limits

• Takes no reference image; it generates from text only
• Does not reliably write dense text such as long paragraphs or tables
• In crowded scenes, headcounts and hand detail can drift

Example prompt

A weathered brass compass resting on a nautical chart, morning window light, shallow depth of field, the words "TRUE NORTH" engraved on the lid, editorial product photography

Example request

POST /v1/queue/modellane/seedream-4.5

Playground

The identifier, price, parameters, and example requests on this page can be read without signing in. Running the model from the panel sends a real generation request and charges the project balance; the playground therefore requires an account.