Seedream 5.0 Pro is ByteDance's flagship image generation and editing model, launched on July 8, 2026. It is not a language model and it does not process video: it takes a text prompt, an optional source image, and up to ten reference images, and returns a still image at 1K or 2K.
Three things separate it from the Seedream 4.x line. It plans a layout before it renders, which is why dense infographics and technical drawings hold together instead of collapsing into decorative noise. It writes on-image text natively in 14 languages, with correct letterforms and reading direction. And it edits a chosen region — a label, a colour, a product — while leaving the rest of the frame untouched.
Is dola-seedream-5-0-pro-260628 the same as Seedream 5.0 Pro?
Yes. ByteDance ships the same model under two IDs. Volcano Engine, its mainland China platform, calls it doubao-seedream-5-0-pro-260628. BytePlus ModelArk, the international platform, calls it dola-seedream-5-0-pro-260628 — "Dola" is BytePlus's international model prefix, not a separate company or a separate model. GPTProto uses the BytePlus ID, so the string you pass to the Seedream 5.0 Pro API here is dola-seedream-5-0-pro-260628.
Seedream 5.0 Pro specifications
| Spec | Value |
|---|---|
| Provider | ByteDance (Seed) |
| Released | July 8, 2026 |
| GPTProto model string | dola-seedream-5-0-pro-260628 |
| Volcano Engine ID (China) | doubao-seedream-5-0-pro-260628 |
| Tasks | text-to-image (this page) · image-edit (reference images) |
| Call pattern | Async — submit a task, poll /api/v3/predictions/{result_id}/result |
| Request fields | prompt, size, enable_base64_output, enable_sync_mode (plus images on image-edit) |
| Output | 1K tier and 2K tier, 9 aspect ratios each |
| Reference images | Up to 10, on the image-edit task |
| On-image text | 14 languages |
| Price on GPTProto | $0.0405 at 1K · $0.081 at 2K — a flat 10% off list |
Seedream 5.0 Pro API Pricing: a Flat 10% Off
GPTProto prices the model at exactly 10% under ByteDance's list rate, at every tier — there is no intro rate that stops applying once you scale up.
| What you call | ByteDance list | Seedream 5.0 Pro API on GPTProto |
|---|---|---|
| 1K image (1024×1024, 1248×832, 1568×672 …) | $0.045 | $0.0405 |
| 2K image (2048×2048, 2848×1600, 3136×1344 …) | $0.09 | $0.081 |
| Each reference image after the first | $0.003 | +$0.0027 |
Two details worth knowing before you budget. Aspect ratio is free: all nine ratios in a tier cost the same, so a 1568×672 banner costs exactly what a 1024×1024 square costs. And the first reference image is included, so a single-reference edit on the image-edit task still bills at the base rate.
A 1,000-image batch at 1K costs $40.50 here against $45 direct. At 2K it is $81 against $90. The per-call saving is small and it compounds per batch — but the reason most teams move is the second column of the ledger: the same balance also covers Seedance, Kling, GPT Image 2, and every other model on the platform, so you are not funding and reconciling a separate wallet per vendor. Full per-size rates are in the pricing table above; account-level rates are on the pricing page.
Seedream 5.0 Pro vs Seedream 5.0 Lite, 4.5 and GPT Image 2
Every model below is live on GPTProto, so the prices are what you actually pay here — not list prices scraped from a vendor page.
| Model | GPTProto model string | Tier / role | Output | Price on GPTProto |
|---|---|---|---|---|
| Seedream 5.0 Pro | dola-seedream-5-0-pro-260628 |
Top tier: layout reasoning, 14-language text | 1K / 2K | $0.0405 (1K) · $0.081 (2K) |
| Seedream 5.0 Lite | seedream-5-0-260128 |
Faster, cheaper drafts; same 5.0 reasoning, lower structural stability | up to 4K | $0.0298 per image |
| Seedream 4.5 | seedream-4-5-251128 |
Previous flagship; strong typography | up to 4K | $0.034 per image |
| Seedream 4.0 | seedream-4-0-250828 |
High-volume batch generation | up to 4K | $0.0255 per image |
| GPT Image 2 | gpt-image-2 |
OpenAI's April 2026 image model; mask-based editing | 2K | $6.40 / 1M input tokens · $24 / 1M output tokens |
Seedream 5.0 Pro vs GPT Image 2 — read the billing unit first.
These two are not priced on the same axis. Seedream 5.0 Pro bills a flat amount per image: $0.0405 at 1K, whatever the prompt. GPT Image 2 bills per token, so a call costs $6.40 per million tokens in and $24 per million out, and your real cost per image moves with output resolution and quality tier. A flat per-image rate is easier to forecast for batch jobs; a token rate can come out lower on small, simple outputs. Both run on one GPTProto key, so the honest answer is to run your own prompt through each and compare the invoice — the GPT Image 2 page has its full rate card.
On capability: pick Seedream 5.0 Pro when the deliverable carries typography, a dense layout, or non-Latin script. Pick GPT Image 2 when you want mask-based edits and conversational instruction-following on a single subject.
Seedream 5.0 Pro vs Seedream 5.0 Lite.
Lite is $0.0107 cheaper per image and reaches 4K, so it is the right default for drafts and volume. Pro is the one to call when the image has to survive a client review: it plans structure before rendering, so multi-element posters, charts, and product scenes come back usable in one pass rather than three.
Switching from BytePlus or Volcano Engine
The model is identical. What changes is who lets you in, and what the request looks like.
- Volcano Engine publishes the list rate but verifies accounts with a mainland China (+86) phone number. Overseas SIMs are rejected.
- BytePlus ModelArk accepts email signup, then gates production access behind a business verification flow.
- GPTProto is an email signup, a top-up, and a key — 10% under the list rate, and no phone or company paperwork.
Be aware of one real difference before you port code: GPTProto serves Seedream 5.0 Pro through an async task API, not a synchronous images endpoint. You POST your prompt to /api/v3/doubao/dola-seedream-5-0-pro-260628/text-to-image, get a result_id back immediately, then poll /api/v3/predictions/{result_id}/result until the image is ready. The model string, the prompt, and the size value all carry over from BytePlus unchanged; the polling loop is the part you add. Copy-paste request and polling code sits in the Quick Start section below.
How to Design With the Seedream 5.0 Pro API
Three jobs this model is built for, with prompts you can paste straight into the playground.
1. Infographic / data poster — where "layout reasoning" actually shows up.
Vertical 9:16 educational infographic, topic "How a heat pump works". Warm ivory background, sage green accents, modern minimalist layout. Four numbered stages with labelled arrows, a small efficiency bar chart bottom-right, clean white space, all labels legible at small size.
2. Localised ad creative — one prompt, one language switch, no re-typesetting.
Product banner for a matcha latte, 16:9, soft studio light on pale linen. Headline text in Japanese reading "夏の新作", subheading in smaller weight below. Preserve Japanese typographic conventions.
3. Photoreal product still at 2K — the $0.081 tier, and the one that doubles as a video keyframe.
Editorial product photograph of a translucent perfume bottle on wet slate, 2048x2048, dramatic side light, hyperreal glass refraction, visible water beading, professional colour grading, shallow depth of field, no text.
Pro tip: put the composition rules first and the style adjectives last. Seedream 5.0 Pro plans structure before it renders, so structural instructions read earlier carry more weight than they would on a diffusion-only model.
