Editorial cover titled Seedream 4.5 API with subtitle Integration Guide, Pricing and Code: an amber key unlocking a glass image-generation pipeline with a 4K spec capsule

The short answer: three ways to call Seedream 4.5 through an API

Yes, Seedream 4.5 has a real API, and no, you do not need a ByteDance account to use it. The official route is BytePlus ModelArk, ByteDance’s international cloud, which bills USD 0.04 per output image (model seedream-4-5-251128, price read from the official model pricing documentation on August 20, 2026). The domestic Chinese route is Volcengine Ark with the same model family. The third route — the one most developers end up on — is an aggregator API: platforms that license Seedream 4.5 and expose it behind one key with no activation step. Aggregator prices on the same day ranged from $0.0325 to $0.0414 per image.

Two definitions before the details, because this SERP mixes them constantly. Seedream is ByteDance’s image generation family (4.0 → 4.5 → 5.0). Seedance is the video model family — a different product with different endpoints; we cover its access separately in our Seedance 2.0 API guide. And “Seedream 4.5” in consumer apps (Dreamina, 即梦) is not the API; this article is about the API only.

Full disclosure up front: Modellix is an aggregator, and we benefit if you pick the aggregator route — we run Modellix, and the Seedream 4.5 model pages we link below are ours. Every price and requirement in this guide was pulled from first-party pages on August 20, 2026, so you can verify each claim — and the honest conclusion includes when the official route is the better trade.

Seedream 4.5 API facts: what you are actually integrating

Seedream 4.5 is ByteDance’s high-fidelity image model, released in early December 2025 as the successor to Seedream 4.0. The official product page leads with three capabilities: reference-image consistency (preserving facial features, lighting, and color tone in multi-image edits), professional editing, and strong typography/dense-text rendering. The official API reference fills in the API-relevant numbers:

Fact Seedream 4.5
Output resolution 2K (default, 2048×2048) or 4K (up to 4096×4096); pixel range 2560×1440–4096×4096
Reference images Up to 14 per request (jpeg/png/webp/bmp/tiff/gif/heic/heif, ≤30 MB each)
Batch / sequential generation Yes — sequential_image_generation: auto, input references + output ≤ 15 images per request
Billing Per successfully generated image (usage.generated_images); content-review rejections are not billed
Watermark watermark: true by default (corner “AI-generated” mark); set false to disable
Output URLs Expire within 24 hours — download before then
Open source? No — API model, no weights released

For positioning, the family lineup matters more than the spec sheet: Seedream 4.0 (API seedream-4-0-250828, $0.03/image) is the previous generation; Seedream 4.5 adds the editing and typography upgrades; Seedream 5.0 series is the newest reasoning-enhanced line. If you are reading this because you searched “seedream api” generically, the 4.5 route below is ByteDance’s current default for text-to-image and image editing workloads.

Seedream 4.5 API pricing: official vs aggregators (August 2026)

Image APIs bill per output image, and Seedream 4.5 pricing is flat per image on every route — no resolution multiplier, no token math on the aggregator side. What varies is the list price and the activation friction around it. All numbers below were read live on August 20, 2026:

Route Price per image Activation required Notes
BytePlus ModelArk (official) $0.04 (2K and 4K) Yes — model activation after funding Input images free; only successful outputs billed
Modellix (aggregator) $0.0414 (T2I and I2I) No — key works immediately See Seedream 4.5 T2I model page and Seedream 4.5 I2I model page
OpenRouter (aggregator) $0.04 No Flat per image, 4K output listed
Kie (aggregator) $0.0325 (6.5 credits) No Publishes “19% below official price”

The official price is documented on the BytePlus model pricing page (output $0.04/image, input free, batch billed by actual generated images). OpenRouter’s Seedream 4.5 page lists $0.04 flat. Kie publishes its credit price on its Seedream 4.5 page.

The honest reading: the official route and the cheapest aggregator differ by about $0.0075 per image — roughly 19%. At 1,000 images that is a $7.50 difference. That gap is real but small; the decision between routes is dominated by activation friction and workflow fit, which is what the next two sections are about. For a single live comparison of every model’s current per-call price, the full list is on our pricing page.

Two-route Seedream 4.5 API pricing diagram: the official BytePlus cloud path behind a gate marked activation, and the aggregator path straight to the model behind one API key

Figure: the two access shapes for Seedream 4.5 — the official cloud gates the model behind activation, the aggregator path does not. Both end at the same underlying model.

Route 1: official BytePlus ModelArk, step by step

The official international path is BytePlus ModelArk (region ap-southeast-1). The step that trips up most developers is not the account — it is model activation: BytePlus requires a funded balance or a purchased resource pack before models can be activated, and the image-generation reference lists model activation under Prerequisites. The video-model tutorial on the same platform states the threshold explicitly (a balance above USD 30 or a purchased resource pack, see Create a video generation task); the console activation flow is the same for image models.

  1. Create a BytePlus account and fund your balance above USD 30, or purchase a resource pack.
  2. Generate an API key in the BytePlus console’s API key management.
  3. Activate the model. Find Seedream 4.5 in model management and activate it — the activation step refuses without step 1.
  4. Call the API. Unlike ByteDance’s video endpoints, image generation is a single synchronous request: POST https://ark.ap-southeast.bytepluses.com/api/v3/images/generations, and the response carries the image directly.

A minimal text-to-image call with curl:

1
2
3
4
5
6
7
8
9
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-4-5-251128",
    "prompt": "A miniature yurt village at night, warm orange light from the windows, dark blue sky",
    "size": "2K",
    "watermark": false
  }'

The same call in Python:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
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-4-5-251128",
        "prompt": "A miniature yurt village at night, warm orange light from the windows, dark blue sky",
        "size": "2K",
        "watermark": False,
    },
    timeout=120,
)
data = resp.json()["data"][0]
print(data["url"])   # expires within 24 hours — download immediately

Parameters that matter for Seedream 4.5 specifically: size accepts 2K or 4K (or exact pixel dimensions), image accepts up to 14 reference images for editing workflows, sequential_image_generation: auto returns a related batch (input + output ≤ 15 images), stream: true streams each image as it finishes, and watermark: false removes the corner mark where your license allows. Read the model ID from configuration — ByteDance version-suffixes these IDs (seedream-4-5-251128, the 251128 being a date), and when the model is updated the old ID can stop resolving. Hard-coding it is how integrations silently break.

For image-to-image work — the “Seedream 4.5 edit” and “image to image” searches — pass reference images in the image field:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import os
import base64
import requests

with open("product_shot.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": "seedream-4-5-251128",
        "prompt": "Swap the background to a clean white studio, keep the product exactly as-is",
        "image": [f"data:image/png;base64,{b64}"],
        "size": "2K",
        "watermark": False,
    },
    timeout=180,
)
print(resp.json()["data"][0]["url"])
Four-stage image generation lifecycle: SUBMIT, GENERATING, RESULT URL, DOWNLOAD chips, with SUCCESS and FAILED statuses and a 24-hour expiry note

Figure: the Seedream 4.5 API call lifecycle — one synchronous request on the official route, one submit-and-poll on aggregator routes. Both end in a download-before-expiry step.

Route 2: an aggregator API — one key, no activation

Aggregator platforms license Seedream 4.5 and resell it behind their own REST endpoints. The trade-off mirrors the official route: no balance top-up, no model activation, per-image billing through a middle layer, and no direct vendor relationship. A Seedream 4.5 API key from an aggregator works the moment you create it. Modellix is one such aggregator (we run it, and we want you to evaluate us on the same terms you would evaluate any provider). Signing up, creating a key, and calling the model is the whole setup:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import time
import requests

BASE = "https://api.modellix.ai/api/v1"
HEADERS = {"Authorization": "Bearer YOUR_MODELLIX_API_KEY", "Content-Type": "application/json"}

# Text-to-image — async submit, then poll
payload = {
    "prompt": "A miniature yurt village at night, warm orange light from the windows, dark blue sky",
    "ratio": "4:3",
    "resolution": "2K",
}
resp = requests.post(f"{BASE}/bytedance/seedream-4.5-t2i/async", headers=HEADERS, json=payload)
data = resp.json()["data"]
print("submitted:", data["task_id"])

for _ in range(120):
    result = requests.get(data["get_result"]["url"], headers=HEADERS).json()
    if result["data"]["status"] in ("success", "failed"):
        break
    time.sleep(2)   # exponential backoff in production

print("status:", result["data"]["status"])

Modellix wraps image generation as an asynchronous task (submit → poll → result), with webhooks available by passing an X-Webhook-URL header — the exact contract is in the Modellix API documentation, and every model’s request schema is published on its model page. The same key unlocks ByteDance’s video models too — the ByteDance provider hub lists the current set, and our guide to running many AI models behind one API key explains the workflow trade-off that motivates it. Treat both code snippets as skeletons, not production modules — verify field names against the current docs for the model ID you actually use.

Seedream 4.5 API Reference

See the full request schema, model IDs, and error codes for Seedream 4.5 on Modellix.

View Docs
Multi-reference Seedream 4.5 editing concept: a grid of reference product images flowing into a generator core that preserves subject identity while changing background and lighting

Figure: what Seedream 4.5’s 14-reference editing is for — keep the subject identical, change everything around it, in one API call.

Official vs aggregator: the honest comparison

The decision is not “who is cheaper” — it is “which wrapper fits your constraints.” Match the columns before you match the numbers:

Dimension Official: BytePlus ModelArk Aggregator (e.g. Modellix)
Price $0.04/image (input free) $0.0325–$0.0414/image, varies by provider
Activation Balance > USD 30 or resource pack, then activate model None — key works immediately
Account setup Cloud account, key, activation steps Sign-up + API key
Billing unit Per successful output image Per output image
Key scope ByteDance/BytePlus models only Many providers behind one key
Output contract Official docs define size, watermark, expiry rules Platform mirrors the model’s contract per model page

Choose the official route when: you are all-in on ByteDance image models at serious volume and want the direct vendor relationship, per-token economics at BytePlus scale, and contractual support. The activation investment is one-time, and the list price is the lowest single-provider option.

Choose an aggregator when: you are evaluating Seedream 4.5 next to other image/video models, you want signup-to-first-image in minutes rather than through funding and activation, or you need one billing relationship across several providers. You pay a small per-image margin and sit one layer away from the vendor.

Neither framing supports a blanket “aggregator is cheaper” claim — the $0.0075 gap versus the cheapest aggregator is real, but activation friction is what usually decides the evaluation. A fuller cost comparison across the whole model landscape lives in our cheapest AI API analysis.

Seedream 4.5 error handling: codes, moderation, and model IDs

Most failed first calls are not model errors — they are integration errors. The ones to know before you hit them:

  • 401 Unauthorized — wrong, expired, or mis-regioned key. BytePlus and Volcengine keys are separate; an aggregator key only works on that aggregator’s endpoints.
  • 403 / model not activated — the classic BytePlus trap: account and key exist, activation was never completed.
  • Insufficient balance — per-image billing draws against the balance; the funded amount that cleared activation is not a free budget.
  • Model ID not found — ByteDance rotates version-suffixed IDs (-251128 style). Read the ID from configuration, and re-check it when you update.
  • Content moderation rejection — the killer detail on the official route: rejected images are not billed (usage.generated_images counts only successes), but the request still fails. In batch mode, a rejected image is skipped and the rest continue; an internal 500 stops the batch.
  • Expired output URL — generated image URLs live 24 hours on BytePlus. Persist the output when it lands, not at the end of your pipeline.

The Modellix API documentation publishes the aggregator-side error table (400/401/402/404/429/500/503, with retry guidance per code), and rate limits scale with your top-up tier — 2 concurrent tasks at the lowest tier up to 100 concurrent at the highest, with RPM limits to match. Key hygiene is the part nobody on this SERP covers: keep keys in environment variables, never commit them, and scope them per service. The snippets in this guide read from os.environ for exactly that reason.

Batch generation and cost control at scale

Seedream 4.5 is one of the few image models with a real batch mode: sequential_image_generation: auto generates a related set in one request (reference images + output ≤ 15), and billing counts only the images actually produced. That changes the cost math from “per request” to “per successful image”:

  • Single image: 1 × $0.04 = $0.04 (official) or 1 × $0.0414 = $0.0414 (Modellix)
  • Batch of 15 (2K, all succeed): 15 × $0.04 = $0.60 official
  • 1,000 images for a product catalog: $40 official, $41.40 via Modellix, $32.50 at the cheapest aggregator list price

Two cost levers worth knowing: failed/rejected images cost nothing on the official route (so a 5% rejection rate costs 5% of your time, not your budget), and input reference images are free on both routes — the 14-reference editing workflows bill only the single output. If you are building a pipeline that generates thousands of variants, budget by successful-image count, add retry-with-backoff around 429s and 5xx, and treat the per-image rate as a moving number: prices drift, and the same model can differ by ~20% across providers on any given day.

Batch cost concept: a stack of generated product images with a per-image unit marker, illustrating 15 images per request and billing per successful output

Figure: the batch math — one request can produce up to 15 images, and the bill is per image that actually succeeds.

Seedream 4.5 vs 4.0 vs 5.0: which API should you integrate?

The Seedream family moves fast, and several ranking pages mix versions. The full current inventory is on the Seedream series hub; here is the API-relevant lineup as of August 2026:

Version API model ID (current) Price Why you would pick it
Seedream 4.0 seedream-4-0-250828 $0.03/image Cheapest; fine for plain text-to-image
Seedream 4.5 (this guide) seedream-4-5-251128 $0.04/image Editing consistency, typography, 14-reference workflows
Seedream 5.0 series seedream-5-0-* / lite $0.035–0.045/image Newest reasoning-enhanced line; check model pages for current pricing

For new integrations, 4.5 is the sensible default: the editing and text-rendering upgrades are the reason it exists, and the $0.01/image premium over 4.0 buys the multi-reference consistency that most production workloads actually need. Choose 4.0 only when you generate plain single images at volume and want the cheapest per-image rate. And keep the video side separate — if the query that got you here was actually about moving images, that is Seedance territory, not Seedream.

Start Generating with Seedream 4.5

Log in to call Seedream 4.5 and 210+ other image, video, and audio models on one Modellix key.

Login

FAQ

Is there a free Seedream 4.5 API?

No free tier exists on the official route — activation itself requires a funded balance or resource pack, and every output image bills at $0.04. Some aggregators offer trial credit for new accounts; check each provider’s current terms. Input reference images are free on both routes, which is the only genuinely free part.

How do I get a Seedream 4.5 API key?

Two paths: official BytePlus (account → fund balance > USD 30 or buy a resource pack → activate model → create API key) or an aggregator like Modellix (sign up, create key, call immediately — no activation).

How much does the Seedream 4.5 API cost?

As of August 20, 2026: $0.04 per image on BytePlus and OpenRouter, $0.0414 per image on Modellix, $0.0325 per image on Kie. Prices are per output image; input images are free; failed or moderation-rejected images are not billed.

Does Seedream 4.5 support 4K output?

Yes — size accepts 2K (default) or 4K (up to 4096×4096), and exact pixel dimensions are supported within the documented range. Price is flat regardless of resolution.

Is Seedream 4.5 open source?

No. It is a closed API model; ByteDance publishes no weights, which is why “seedream 4.5 huggingface” searches find re-uploads or third-party wrappers rather than official checkpoints.

Can I use Seedream 4.5 with ComfyUI?

Not through the official API — ComfyUI integration would run through third-party nodes that call an API provider under the hood, not native weights. If your workflow is API-driven, the direct routes in this guide are simpler.

What is the difference between Seedream and Seedance?

Seedream is ByteDance’s image generation family (text-to-image and image editing, $0.04/image at 4.5). Seedance is the video generation family (per-second billing, async task endpoints). They share a vendor and a naming prefix, not an API surface — see our Seedance 2.0 API access guide for the video side.

Can I use Seedream 4.5 API from the USA?

Yes — BytePlus ModelArk is the international route (region ap-southeast-1) and needs no China-region account. Volcengine Ark is the domestic alternative for infrastructure inside mainland China.

Prices, model IDs, and activation rules reflect public information as of August 20, 2026 and change frequently — model IDs rotate on update, and per-image rates drift. Validate against each provider’s live pages before committing. Access Seedream 4.5 and 210+ other image, video, and audio models through a single API key at www.modellix.ai.