A Seedream API key is a ModelArk API key from BytePlus, ByteDance's cloud for international users (or from Volcengine Ark inside mainland China). You create it under API Keys in the ModelArk console, activate the Seedream model you want on the Model activation page, and send it as a Bearer token to https://ark.ap-southeast.bytepluses.com/api/v3/images/generations with a model ID such as dola-seedream-5-0-pro-260628. As of 2 November 2026, BytePlus charges $0.045 per Seedream 5.0 Pro image at 1.5K or below, $0.018 for 5.0 Flash and $0.035 for 5.0 Lite.

Every BytePlus and Volcengine detail below comes from their official documentation, read on 2 November 2026 and linked at the end. If you only want to make images with Seedream and do not need code, skip to the no-key option.

Is there a separate "Seedream API key"?

No. Seedream does not have its own key. ByteDance serves its Seed models (Seedream for images, Seedance for video, Seed for text) through one platform, and one key works for all of them once each model is activated:

  • BytePlus ModelArk is the international version. Its quick-start uses the base URL https://ark.ap-southeast.bytepluses.com/api/v3 (region ap-southeast-1).
  • Volcengine Ark (火山方舟) is the mainland-China version, with its own console, account and keys.

A key from one does not work on the other. Pick BytePlus unless you are building for users in mainland China.

How do you get a Seedream API key on BytePlus?

The steps below follow BytePlus's ModelArk quick start, as of 2 November 2026.

  1. Create a BytePlus account and sign in to the ModelArk console.
  2. Choose the project. Switch to the target project in the lower-left corner of the console.
  3. Create the key. Click API Keys in the left navigation and create a ModelArk API key. Copy it somewhere safe.
  4. Activate Seedream. Open the Model activation page and activate each Seedream model you plan to call (for example Seedream 5.0 Pro and 5.0 Flash). An unactivated model will not answer, even with a valid key.
  5. Store the key as an environment variable. BytePlus's samples read it from ARK_API_KEY: on macOS or Linux, export ARK_API_KEY="your_api_key_here".
  6. Install an SDK (optional). The official Python SDK is arkruntime (python -m pip install --upgrade arkruntime). The docs also show the OpenAI SDK pointed at the ModelArk base URL.
  7. Make your first call (example below).

What does a first Seedream API call look like?

This is BytePlus's own text-to-image sample from the image generation tutorial, shortened:

bash
curl https://ark.ap-southeast.bytepluses.com/api/v3/images/generations \ -H "Content-Type: application/json" \ -H "Authorization: Bearer $ARK_API_KEY" \ -d '{ "model": "seedream-5-0-lite-260128", "prompt": "Vibrant close-up editorial portrait, dramatic studio lighting", "size": "2K", "output_format": "png", "watermark": true }'

To edit an image instead, add an "image" field with a URL (or a list of URLs for multi-image blending). To use another version, change only the model value.

How do you get a Seedream API key in China (Volcengine)?

Volcengine's Chinese guide, 获取 API Key 并配置, describes the same flow: sign in to the Ark console, optionally switch project space in the lower-left corner, open API Key 管理 (API key management), click 创建 API Key (create API key), confirm the name and create it. Volcengine also recommends keeping the key in the ARK_API_KEY environment variable rather than in code. Its key-management page says one main account can create up to 50 API keys, and that a key can be limited to specific projects, model IDs or inference endpoints and to an IP allowlist.

Which Seedream model IDs can you call?

From BytePlus's model list and image generation tutorial, as of 2 November 2026:

ModelModel IDResolutionsNotable abilities
Seedream 5.0 Prodola-seedream-5-0-pro-2606281K, 1.5K, 2KInteractive (marked-region) editing, layer decomposition (1 base + up to 16 layers), multi-reference
Seedream 5.0 Flashdola-seedream-5-0-flash-2609151K, 1.5K, 2KSame editing and layers as Pro, faster and cheaper
Seedream 5.0 Liteseedream-5-0-260128 (also seedream-5-0-lite-260128)2K, 3K, 4KBatch (sequential) image sets, streaming output
Seedream 4.5seedream-4-5-2511282K, 4KBatch image sets, multi-reference
Seedream 4.0seedream-4-0-2508281K, 2K, 4KBatch image sets, multi-reference

The dated suffix matters: the API expects the full ID, not "seedream-5". BytePlus's tutorial also notes that 5.0 Pro and 5.0 Flash do not support batch image output, while Lite allows input reference images plus generated images up to 15 in total.

How much does the Seedream API cost?

BytePlus prices images per output image. From its ModelArk pricing page, as of 2 November 2026:

Model IDPrice per output imageInput images
dola-seedream-5-0-pro-260628$0.045 up to 2.61 MP (1.5K or lower); $0.09 above 1.5KFirst image free, then $0.003 each
dola-seedream-5-0-flash-260915$0.018Free
seedream-5-0-lite-260128$0.035Free
seedream-4-5-251128$0.04Free
seedream-4-0-250828$0.03Free

Layer decomposition on 5.0 Pro is billed per layer ($0.0225 at 1.5K or lower, $0.045 above). Two billing rules are worth knowing: images that fail because of content moderation are not billed, and batch generations on Lite, 4.5 and 4.0 are billed by the number of images actually generated. The pricing page also warns that an account left in arrears for 2 hours has its model services suspended until it is topped up.

Is there a free Seedream API?

There is a free trial, not a free tier. BytePlus's free trial page lists a free quota of 200 images for Seedream 4.5 for new, eligible BytePlus users, and says the free quota does not expire but is subject to change. After that, every image is paid at the rates above. The guide to using Seedream for free covers the other no-cost routes.

What are the Seedream API rate limits?

BytePlus lists a maximum of 500 images per minute (IPM) for every Seedream model, from 4.0 to 5.0 Pro, on its model list. It also says these are theoretical maximums that depend on platform load, that you can see your account's limits on the Model activation page, and that you can request a higher limit from the console. Region matters too: all listed models run in ap-southeast-1, while seedream-5-0-lite is also available in eu-west-1 (https://ark.eu-west.bytepluses.com/api/v3).

How do you keep a Seedream API key safe?

  • Never put the key in front-end code or a public repository. Call the API from your server and read the key from ARK_API_KEY.
  • Use one key per app or environment, so you can disable a leaked key without breaking everything else. Volcengine's docs note that requests with a disabled key fail authentication.
  • Restrict keys to the projects and models they need (Volcengine supports model-ID and IP allowlist limits).
  • Watch the balance. Because service stops after 2 hours in arrears, set a low-balance alert if your app depends on it.

What do Seedream outputs look like?

Two real Seedream 5.0 Pro images we made on Kenerate (which calls the same ByteDance model). Prompts are the exact ones used.

Ukiyo-e style illustration of a geisha in a crimson kimono with a paper umbrella under cherry blossoms at night, made with Seedream 5.0 Pro
Ukiyo-e style illustration of a geisha in a crimson kimono with a paper umbrella under cherry blossoms at night, made with Seedream 5.0 Pro

Seedream 5.0 Pro, 2K. Prompt: "Elegant geisha in a crimson silk kimono holding a paper umbrella under cherry blossoms at night, ukiyo-e inspired illustration, soft lantern glow."

Flat paper-cut style poster of a mountain lake at sunset with a red canoe, made with Seedream 5.0 Pro
Flat paper-cut style poster of a mountain lake at sunset with a red canoe, made with Seedream 5.0 Pro

Seedream 5.0 Pro, 1K. Prompt: "Flat 2D wall art poster: a minimalist mountain lake at sunset with layered paper-cut style mountains, a small red canoe, two birds, limited palette of teal, coral and cream, subtle grain texture, mid-century modern illustration, no text."

For prompt structure that suits the API as well, see the Seedream 5.0 prompt guide.

What about third-party Seedream APIs?

Several model aggregators resell Seedream behind their own API keys and their own prices. They can be convenient if you already use them for other models, but you are then bound by their pricing, model versions and terms, not ByteDance's. Compare their per-image price with the BytePlus table above before choosing, and check which exact Seedream model ID they route to.

Can you use Seedream without an API key?

Yes. Kenerate AI runs the Seedream family in the browser with credits, so there is no key, console, model activation or billing setup. As of 2 November 2026 (from Kenerate's models-config.ts and image-credit-calculator.ts):

Seedream on KenerateCredits per imageReference images
Seedream 5.0 Pro13.5 at 1K/1.5K, 27 at 2KUp to 10
Seedream 5.0 Lite10.5 (2K to 4K)Up to 10
Seedream 5.0 Flash8.1Up to 10
Seedream 4.512Up to 10
Seedream 4.08.1Up to 10

Credits come in one-time packs from $15 for 1,300 credits, which never expire. Being direct about the maths: one credit costs about 1.15 cents on the $15 pack (and 1 cent on the $109 pack of 10,900), so a 13.5-credit Seedream 5.0 Pro image costs about $0.156 (or $0.135), against $0.045 on the BytePlus API. If you are generating thousands of images in code, the direct API is cheaper. Kenerate makes more sense when you want no setup, a few images at a time, or Seedream next to GPT Image, Nano Banana and video models under one balance.

You can also call Seedream from an AI assistant without a ByteDance key through the Kenerate MCP server, which uses your Kenerate credits. Model details and more real outputs are on the Seedream 5.0 page, the Seedream 4.5 page and the Seedream hub.

Which route should you choose?

  • Building an app or batch pipeline: BytePlus ModelArk API key, dola-seedream-5-0-flash-260915 for volume, dola-seedream-5-0-pro-260628 for quality and edits.
  • Serving users in mainland China: a Volcengine Ark key.
  • Trying Seedream before paying anything: the BytePlus free trial (200 Seedream 4.5 images for eligible new users) or Kenerate's 25 sign-up credits.
  • No code at all: Seedream in Kenerate's Image Studio or through an AI assistant via MCP.