ByteDanceByteDance image model

Seedream 4.5. Production resolution, precise edits.

ByteDance's flagship image model generates at resolutions ready for print and large-format use, edits existing images with precision, and composes scenes from multiple references.

  • Text and image input
  • Print-ready resolution
The model

Seedream 4.5 is built for output that leaves the screen. Generations come back at high resolution with detail to match, so crops, print, and large-format placements work from the raw asset instead of an upscale. It takes image input too, handling targeted edits and multi-reference composition: this product, that background, this style, one coherent frame.

In Treza pipelines Seedream 4.5 pairs naturally with its video sibling: art direct a key frame in Seedream, then hand it to Seedance 2.0 as a keyframe or reference and the campaign's stills and motion stay in one visual family. Like everything on the canvas, it publishes behind the same API.

High-resolution output

Native generations with enough pixels for print, banners, and aggressive crops, no upscaling stage required.

Precise editing

Feed an image and an instruction for targeted changes that respect the rest of the frame.

Multi-reference composition

Combine product, style, and scene references into one image that follows all of them.

Pairs with Seedance video

Stills from Seedream drop into Seedance 2.0 as keyframes and references, keeping campaigns consistent across media.

Prompt to result

What to make with Seedream 4.5

Real prompts, and what comes back. Every one runs in Treza chat or as a node on the canvas.

A key visual for a festival poster, A2 print, dense detail

A print-resolution asset straight from the model.

This handbag (attached) in the style of this reference photo (attached)

A composition following both references at once.

Replace the sky in this shot with dramatic storm clouds

A targeted edit with the foreground untouched.

A hero banner at 21:9 of a smartwatch over a topographic map

A wide-format asset that stays sharp edge to edge.

Generate the opening frame for a Seedance clip: mascot at a red door

A key frame ready to hand to the video node.

Four seasonal variants of this storefront scene

A consistent set from one master composition.

Beyond one prompt

Seedream 4.5 is one node on a canvas

Chain it with language, image, and video models, then publish the whole pipeline as an API.

  1. Step 01

    Draft the brief

    A language model node expands a one-line idea into a detailed, on-brand prompt before it ever reaches the model.

  2. Step 02

    Generate with Seedream 4.5

    Run variants in parallel, compare against other models on the same prompt, and keep every result in run history.

  3. Step 03

    Publish it as an API

    The pipeline becomes a versioned endpoint. Your product calls it with the typed /invoke API or any OpenAI SDK.

FAQ

Seedream 4.5, answered

How high resolution are Seedream 4.5 generations?

High enough for print and large-format use straight from the model, which is its core distinction from flash-tier generators. If your last step used to be an upscaler, Seedream usually removes it.

Can Seedream 4.5 edit images and combine references?

Yes. It accepts image input for targeted edits, and it can compose from multiple references at once: a product photo plus a style reference plus a scene, resolved into one coherent image.

How does Seedream 4.5 work with Seedance 2.0?

They are siblings from ByteDance and share an aesthetic. A common Treza pipeline art-directs a still in Seedream 4.5, then passes it to Seedance 2.0 as a first-frame keyframe or subject reference, so stills and video in a campaign match.

How much does Seedream 4.5 cost on Treza?

Treza uses prepaid credits with no subscription. Each image is charged at Seedream 4.5's rate from your balance, and only successful generations are charged. Credit packs start at $5 and never expire.

Can I call Seedream 4.5 through an API?

Yes. Build a pipeline that uses Seedream 4.5 and publish it as a versioned HTTP endpoint. Call the typed /invoke API from your product, or point any OpenAI SDK at the OpenAI-compatible endpoint. If a better model ships later, swap it on the node without changing the API your product calls.