Seedream 5.0 API: Integration Guide, Pricing & Code

Yes, the Seedream 5 API exists, it is live, and it bills in dollars you can look up today. “Seedream 5” is the shorthand developers use for Seedream 5.0, ByteDance’s current image-generation family — two tiers, Lite and Pro, both for sale on the official cloud and both callable through third-party aggregators. As of September 7, 2026, the official international route (BytePlus ModelArk) lists Lite at $0.035 per generated image under model ID seedream-5-0-lite-260128, and Pro at $0.045 per image for output at 1.5K or lower (double above that) under the ID printed in BytePlus’s own pricing table. This guide gives you the four things the ranking pages scatter across a dozen tabs: the naming straight, the official pricing pulled today, working request code, and an honest answer to which tier and which route fit your workload. Full disclosure up front: Modellix publishes this blog and sells API access to Seedream 5.0 — we state our prices, compare them against the official list, and flag where the official route is cheaper.

Seedream 5 API is Seedream 5.0: naming, tiers, and what “current” means

If you searched “seedream 5 api,” you are looking at the Seedream 5.0 generation. ByteDance’s image family runs Seedream 4.0 → 4.5 → 5.0, and version five ships as two API products rather than one model:

• Seedream 5.0 Lite — the workhorse: text-to-image, image-to-image, editing, sequential batches, and output up to 4K. ByteDance’s Lite product page describes it as a unified multimodal model with deep thinking and online-search capabilities. In the API it is the tier that supports batch generation and streaming.
• Seedream 5.0 Pro — the design-understanding flagship, added to the official catalog in mid-2026 (its model-ID date suffix is 260628): layer decomposition into up to 16 editable layers, coordinate-level interactive editing, native text in more languages, single-image generation up to 2K. See the Pro product page for the marketing view.

Three naming traps trip up most first integrations, and this SERP mixes all of them:

1. 5 vs 5.0: the official name is Seedream 5.0. Marketplaces and blog posts that write “Seedream 5 API” (including the one you are reading) mean the same family.
2. Seedream vs Seedance: Seedream is the image family. Seedance is ByteDance’s video family — different endpoints, per-second billing. If your project generates video, that is a different API entirely.
3. Model IDs differ by region and tier: the international BytePlus route prints Pro as dola-seedream-5-0-pro-260628 (BytePlus markets the international Pro as “Dola” — its own announcement says “Dola Seedream 5.0 Pro API is now available on BytePlus”), while the domestic Chinese route (Volcengine) uses doubao-seedream-5-0-pro-260628; Lite is seedream-5-0-lite-260128 internationally and doubao-seedream-5-0-lite-260128 domestically. Copy the ID from your own console rather than from a blog — ByteDance date-suffixes these IDs and rotates them on releases.

Lite arrived in early 2026 and Pro followed mid-2026; both are current-generation today. If you hold a 4.0 or 4.5 pipeline, the “should you move” question is answered in the second-to-last section.

What the Seedream 5.0 API does: Lite vs Pro capabilities

Both tiers share one API surface — the official image-generation API reference (last updated August 12, 2026) documents Seedream 5.0 Pro, 5.0 Lite, 4.5, and 4.0 against the same endpoint. The split that matters is capability, not quality:

Capability	Seedream 5.0 Lite	Seedream 5.0 Pro
Text-to-image / image-to-image / editing	✅	✅
Reference images per request	Up to 14	2–10
Sequential batch (sequential_image_generation: auto)	✅ up to 15 images	❌ not supported
Streaming output (stream: true)	✅	❌ not supported
Layer decomposition (≤16 layers, PNG + alpha)	❌	✅
Interactive editing (coordinates / boxes / arrows)	❌	✅
Resolution levels	2K / 3K / 4K (explicit pixels up to 4096×4096)	1K / 1.5K / 2K
Prompt languages (API docs)	Chinese + English	17 languages incl. Arabic, Thai, Korean, Spanish, French, German, Vietnamese, Japanese

Figure: the Seedream 5.0 lineup — one API core, two tiers with different batch, resolution, and layer capabilities. Concept diagram generated for this article, not a screenshot.

The practical reading: Lite is the throughput tier (related batches, streaming, larger output canvas); Pro is the precision tier (editing control and layered output for production design pipelines). ByteDance’s reference also recommends keeping prompts under roughly 300 Chinese characters or 600 English words across the family.

Seedream 5.0 API pricing, pulled September 7, 2026

The authoritative numbers come from BytePlus’s model pricing page, read today. BytePlus bills per successfully generated image — failed and moderation-rejected outputs are not charged — and input images are free on Lite. Pro prices are split by an output-pixel threshold: ≤ 2.61 million pixels (1.5K or lower) vs above it.

Route (checked 2026-09-07)	Seedream 5.0 Lite	Seedream 5.0 Pro (single image)	Input images	Activation
BytePlus ModelArk (official)	$0.035/image	$0.045 (≤2.61MP / 1.5K) · $0.09 (>2.61MP); layer decomposition $0.0225 / $0.045	Free (Lite) · Pro: 1st free, then $0.003 each	Funded account + model activation
Modellix (aggregator)	$0.0350/image	$0.0810/image (flat, as listed on the Lite and Pro model pages)	Free	None — key works immediately
fal.ai (aggregator)	not quoted here	from $0.0675 (≤1536×1536; $0.135 up to 2048×2048, per fal’s Seedream 5.0 page)	—	None
kie.ai (aggregator)	≈ $0.0275 (5.5 credits, ~21% below official per kie’s Lite page)	separate Pro page	—	None; trial credits for new accounts

Two honesty notes on that table. First, this is a same-model, same-task comparison — it is not a claim that aggregators are generally cheaper. On Lite, Modellix lists the official price flat ($0.0350; re-verified against the live Lite model page on September 8, 2026, in both the server render and the JavaScript render) and kie undercuts both; on Pro, the official route’s 1.5K-and-below tier ($0.045) is cheaper than every aggregator flat rate we checked, because BytePlus prices small Pro output aggressively. Second, “from $0.0675” on fal is a starting price that scales with output size — the fair comparison for Pro is the $0.09 tier against fal’s $0.135 at 2048×2048, where aggregators narrow the gap. ByteDance’s international product page (byteplus.com/en/product/Seedream) summarizes the same structure: “starting at $0.045/image, first reference image free, additional from $0.003.” Volcengine, the domestic route, prices separately in local currency — if your infrastructure runs inside mainland China that is the page to read, not this table.

Doing the arithmetic on the official route: a 1,000-image catalog at Lite is $35; at Pro’s low tier it is $45 (or $90 if you push every output above 1.5K); layer-decomposing 1,000 images at the low tier is $22.50 plus the free first input image each.

The official route: BytePlus ModelArk, step by step

BytePlus ModelArk is ByteDance’s international cloud — no China-region account required. Setup has two steps people underestimate, so the order matters:

1. Create a BytePlus account and fund it. Model activation is gated behind a funded balance or purchased resource pack; BytePlus’s tutorials quote the threshold differently over time (commonly “above USD 30”), so read the current figure in your console.
2. Create an API key under API-key management. It is shown once — store it in a secrets manager, never in code.
3. Activate Seedream 5.0. Find the model in model management (Lite and Pro are separate activations) and activate it. This step refuses until step 1 is satisfied — the classic “my key is valid, why 401” trap.
4. Call the endpoint. Unlike ByteDance’s video models, image generation is a single synchronous request: one POST returns the image URL(s) in the response.

The endpoint, from the official API reference: POST {base}/images/generations where the base URL depends on region — ap-southeast-1 at https://ark.ap-southeast.bytepluses.com/api/v3 or eu-west-1 at https://ark.eu-west.bytepluses.com/api/v3 for European workloads and data residency. The same reference notes that BytePlus exposes rate-limit and quota controls through model units / Endpoint IDs rather than a fixed published number — create an Endpoint ID if you need guaranteed concurrency, and read your limit from the console. For the domestic route, Volcengine documents the same family under its doubao model IDs (e.g., doubao-seedream-5-0-pro-260628).

Working code: curl, Python, and a layer-decomposition call

curl — text-to-image on Lite. The official route takes a bearer key and a JSON body with model, prompt, and size:

curl https://ark.ap-southeast.bytepluses.com/api/v3/images/generations \
  -H "Authorization: Bearer $ARK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedream-5-0-lite-260128",
    "prompt": "A miniature tea house at night, warm lantern light, deep blue snow-covered garden",
    "size": "2K"
  }'

Python — the same call with requests, reading the model ID from configuration (never hard-code it — ByteDance rotates date-suffixed IDs on releases):

import os
import requests

resp = requests.post(
    "https://ark.ap-southeast.bytepluses.com/api/v3/images/generations",
    headers={"Authorization": f"Bearer {os.environ['ARK_API_KEY']}"},
    json={
        "model": "seedream-5-0-lite-260128",
        "prompt": "A miniature tea house at night, warm lantern light, deep blue snow-covered garden",
        "size": "2K",
    },
    timeout=120,
)
data = resp.json()["data"][0]
print(data["url"])   # expires within 24 hours — download it now, not later

Python — Seedream 5.0 Pro layer decomposition, which is where the generation-comparison becomes concrete. Pro takes one input image plus layer_decomposition: true and returns a base image plus named, positioned layers:

import os
import base64
import requests

with open("poster.png", "rb") as f:
    b64 = base64.b64encode(f.read()).decode()

resp = requests.post(
    "https://ark.ap-southeast.bytepluses.com/api/v3/images/generations",
    headers={"Authorization": f"Bearer {os.environ['ARK_API_KEY']}"},
    json={
        "model": "dola-seedream-5-0-pro-260628",   # as printed in BytePlus docs; confirm in your console
        "prompt": "Separate the headline text, the product, and the background into layers",
        "image": f"data:image/png;base64,{b64}",
        "layer_decomposition": True,
        "size": "2K",
        "response_format": "url",
    },
    timeout=180,
)

for item in resp.json()["data"]:
    print(item.get("z_index"), item.get("name"), item.get("url"),
          item.get("bounding_box", {}).get("absolute"))

Response fields worth knowing: data[].url expires 24 hours after generation (persist immediately); response_format can return url or b64_json; size on Lite accepts 2K/3K/4K or explicit pixels, on Pro 1K/1.5K/2K; watermark: false is available on supported models; and in layer decomposition every layer is PNG with an alpha channel, with z_index, name, and bounding_box telling you how to reassemble the design. For editing (not decomposition), pass one or more reference images in the image field with an instruction prompt — Lite accepts up to 14 references, Pro up to 10.

Figure: the request/response lifecycle for Seedream 5.0 — synchronous on the official route, billed per successfully generated image, output URLs expiring after 24 hours. Concept diagram generated for this article.

The aggregator route we describe next uses the same mental model with one difference: tasks are asynchronous — you submit, poll, and retrieve.

The aggregator route: one key for Seedream 5.0 (and what it costs)

If the official route’s funding-and-activation dance is friction you do not want, aggregators license Seedream 5.0 and expose it behind their own key. Modellix — the company publishing this guide — currently lists the full Seedream 5.0 line: Lite, Lite Edit, Pro, Pro Edit, and Pro Multi-Reference, all documented on the ByteDance provider page. Two honest observations about that catalog: first, Modellix lists Lite at the official price ($0.0350) and sets Pro at a flat $0.0810 that sits between BytePlus’s tiers — cheaper than the $0.09 high tier, pricier than the $0.045 low tier, and if you only ever generate at 1.5K-or-below, the official route is the cheaper Pro door. Second, the older Seedream pages (4.0 and 4.5) return “Model Not Found” on Modellix today — checked September 7, 2026 — because the catalog rotates to the current generation. That is the normal lifecycle of this family on marketplaces, and it is exactly why this article exists.

The API shape is a submit-and-poll task: send the model ID and prompt (uploading reference images first through the File API, which is not billed and retains files about 7 days), then poll the task result or register a webhook to receive output URLs. Request logs record each call’s cost, which keeps per-image spend auditable across the 200+ models behind the same key. You trade one contract and one invoice for the fact that you are one step removed from ByteDance’s own infrastructure — for evaluation workloads and multi-model products that is usually the right trade.

Seedream 5.0 on the Modellix REST API

See the task lifecycle, model IDs, and per-call pricing for Seedream 5.0 Lite and Pro in the API reference.

View Docs

Official route or aggregator: which fits whom

If your main constraint is	Start with	Because
Lowest possible Pro cost at ≤1.5K output	BytePlus official	$0.045 beats every aggregator flat rate we checked
Zero account friction today	Aggregator (e.g., Modellix)	No funding threshold, no activation step, key works immediately
One key across many models / vendors	Aggregator	Lite through Pro Edit plus 200+ other image, video, audio, and LLM models on one contract
Europe data residency	BytePlus eu-west-1	Official regional endpoint
Mainland-China infrastructure	Volcengine	Domestic route, local currency
Per-call cost auditability	Either — with different mechanisms	BytePlus usage fields vs aggregator request logs

The honest summary is not “aggregator beats official” — it is that the two routes price the same model with different shapes. Official BytePlus gives you the pixel-tiered low end and direct vendor support but charges setup attention (funding, activation, per-model IDs, regional keys). An aggregator removes all of that and adds multi-model consolidation, and you pay for it either as a flat Pro rate above BytePlus’s low tier or as a small per-image spread on some models. Run a 100-image test on the route you lean toward before committing a pipeline — both are pay-as-you-go, so the test costs single-digit dollars.

Figure: the route decision in one picture — heavier setup at the official cloud, one key through an aggregator, both reaching the same Seedream 5.0 model. Concept diagram generated for this article.

Which Seedream should you integrate in 2026?

The generation question, answered straight:

1. New integration → Seedream 5.0. Everything ByteDance is shipping — layer decomposition, interactive editing, the multilingual text work — lands on the 5.0 line. Our Seedream 4.5 API guide covers the previous generation’s workflow, and the Seedream series hub maps the full family; the Seedream 4.0 price analysis shows how far list pricing has moved.
2. Existing 4.x pipeline with outputs you cannot disturb → stay, but plan the move. Official BytePlus still sells 4.0 at $0.03 and 4.5 at $0.04 (same pricing page as above), and third-party catalogs are where the old generations disappear first — we verified our own 4.0/4.5 pages de-listed. A marketplace-dependent 4.x integration has a clock on it; read model IDs from configuration and budget for rotation.
3. Your query is about video → that is Seedance, a different API with per-second billing; our Seedance 2.0 API access guide covers it. And if you are weighing Seedream 5.0 against non-ByteDance image models, Nano Banana vs Seedream does that comparison without the hype.

Generate with Seedream 5.0

Log in to Modellix to call Seedream 5.0 Lite and Pro alongside 200+ other models through one API key.

Login

Frequently Asked Questions

Is the Seedream 5 API free?

No official free tier exists: every successfully generated image bills (Lite $0.035, Pro from $0.045 on the official route, checked September 7, 2026), and official activation requires a funded account. Some aggregators offer trial credit for new accounts — kie advertises free testing credits — but signup offers change: Modellix’s automatic $1 signup credit ended in August 2026 and trial credit is now granted on request, so check each provider’s live offer. Input prompts are free, which is the only genuinely free part.

How do I get a Seedream 5 API key?

On the official route: fund a BytePlus account, activate the model, and create a key under API-key management; send it as a Bearer token to https://ark.ap-southeast.bytepluses.com/api/v3/images/generations. On an aggregator route you create that provider’s key — one key covers all its models, no ByteDance account needed.

What is the difference between Seedream 5 and Seedream 5.0?

Nothing — “Seedream 5” is the shorthand; Seedream 5.0 is the official name of the current generation (Lite and Pro tiers). Don’t confuse either with Seedance, the video family.

Should I use Seedream 5.0 Lite or Pro?

Lite if you need batch/sequential generation, streaming, 4K output, or the lowest per-image cost ($0.035). Pro if you need layer decomposition, coordinate-level editing, 10-language-plus text rendering, and are willing to pay per-pixel-tier rates ($0.045–$0.09 official).

Can I call Seedream 5.0 from the US?

Yes. BytePlus ModelArk is the international route (ap-southeast-1, plus eu-west-1) and requires no China-region account; Volcengine is the domestic alternative for infrastructure inside mainland China.

Is Seedream 5.0 available on Replicate / other marketplaces?

Marketplaces license it under their own model IDs and prices (fal sells it as bytedance/seedream/v5/..., kie as credit-based listings; Replicate hosts a bytedance/seedream-5 endpoint). Official model IDs from BytePlus’s own docs are the ones in this article — aggregator IDs are per-provider by design.

What happens if a generated image is rejected?

You are not billed: BytePlus charges only successfully generated images, and in sequential mode a moderation-rejected image is skipped while the rest of the batch continues (an internal 500 stops the batch). The response’s usage.generated_images field counts only successes.

Are Seedream 5.0 prices stable?

ByteDance has rotated model IDs and marketplace listings within a single quarter this year, and aggregator rates move independently — the numbers here were re-pulled September 7, 2026, and the model IDs carry date suffixes for a reason. Re-validate against the official pricing page and your provider’s model page before committing budget.

---

Seedream model availability and pricing reflect public information as of September 7, 2026 and change frequently — ByteDance has rotated model IDs and marketplace listings within a single quarter, and aggregator prices move independently. Re-validate against BytePlus’s official pricing page and each provider’s live model page before committing budget. This article was written by Modellix, an API aggregator with a commercial interest in Seedream 5.0 — the official-route integration material above is vendor-neutral, we state plainly where the official route is cheaper than our own listing, and we note that our catalog now serves the 5.0 line only. Access 210+ image, video, and audio models through one API key at www.modellix.ai.