@xieyongliang

BytePlus ModelArk

OpenClaw plugin for BytePlus ARK API — Seed/Kimi/GLM language models, Seedream image generation, and Seedance video generation

当前版本
v2026.4.11
code-plugin社区source-linked

openclaw-plugin-byteplus-modelark

OpenClaw plugin for the BytePlus ARK API — language models, image generation, and video generation.

CapabilityModels
Language modelsSeed 2.0 Pro / Lite, Seed 1.8, Kimi K2.5, GLM 4.7, Ark Code
Image generationSeedream 5.0, 4.5, 4.0
Video generationSeedance 1.5 Pro, 2.0, 2.0 Fast, 1.0 Pro, 1.0 Lite

All capabilities share a single BYTEPLUS_API_KEY.


Installation

openclaw plugins install xieyongliang-byteplus-modelark

Restart the OpenClaw gateway (via the Mac app, or openclaw gateway run) after installation.

For local development:

git clone https://github.com/xieyongliang/openclaw-plugin-byteplus-modelark.git
cd openclaw-plugin-byteplus-modelark
npm install
openclaw plugins install --link .

Configuration

Get your API key from the BytePlus ARK Console.

Option A — Onboarding wizard (recommended)

openclaw onboard
# → Select "BytePlus ModelArk" → enter your BytePlus ARK API key

This writes models.providers.byteplus-modelark.apiKey to your config.

Option B — Plugin config

# API key (all capabilities read this first)
openclaw config set plugins.entries.@xieyongliang/byteplus-modelark.config.apiKey YOUR_API_KEY

# Override base URL (for China region)
openclaw config set plugins.entries.@xieyongliang/byteplus-modelark.config.baseUrl https://ark.cn-beijing.bytedance.com/api/v3

Option C — Environment variable

export BYTEPLUS_API_KEY="your-api-key"
# Optional: override base URL
export BYTEPLUS_BASE_URL="https://ark.cn-beijing.bytedance.com/api/v3"

Key resolution order

For all three capabilities (LLM / Seedream / Seedance), the API key is resolved in this order:

  1. plugins.entries.@xieyongliang/byteplus-modelark.config.apiKey (plugin config)
  2. models.providers.byteplus-modelark.apiKey (set by openclaw onboard)
  3. BYTEPLUS_API_KEY environment variable

Language Models

Supported models

Provider byteplus-modelark (general)

Model IDNameModalityContext
seed-2-0-pro-260328Seed 2.0 Pro (default)text + image256K
seed-2-0-lite-260228Seed 2.0 Litetext + image256K
seed-1-8-251228Seed 1.8text + image256K
kimi-k2-5Kimi K2.5text131K
glm-4-7GLM 4.7text + image128K

Provider byteplus-modelark-plan (coding/planning)

Model IDNameModalityContext
ark-code-latestArk Code (Latest)text256K

Set default model

# Set as your global default
openclaw config set models.default byteplus-modelark/seed-2-0-pro-260328

# Or in OpenClaw web UI: Settings → Models → Default Model

Usage

Command line:

openclaw agent --model byteplus-modelark/seed-2-0-pro-260328 \
  --message "Summarize the key points of microservices architecture"

In a conversation (TUI or web):

What is the capital of France?
Using seed-2-0-lite, summarize this article in 3 bullet points: ...

Deep thinking (Seed models):

--thinking high: Prove that there are infinitely many prime numbers
--thinking medium: Analyze the pros and cons of this architecture diagram

Image Generation (Seedream)

Supported models

Model IDDescription
byteplus/seedream-5-0-260128Seedream 5.0 (default)
byteplus/seedream-4-5-251128Seedream 4.5
byteplus/seedream-4-0-250828Seedream 4.0

Parameters

ParameterTypeDefaultDescription
promptstringText description of the image
modelstringseedream-5-0-260128Model ID
resolutionstring2K1K, 2K, or 4K
seednumberrandomFixed seed for reproducible results
watermarkbooleanfalseAdd watermark to output
guidanceScalenumberPrompt adherence strength (1.0–20.0)
optimizePromptbooleanfalseAuto-enhance prompt before generation
inputImagesimage[]Reference images for image-based generation

Usage

Command line:

openclaw agent --message "Generate a photorealistic image of a misty mountain lake at sunrise"

With parameters (in conversation):

Generate a 4K image of a futuristic cityscape at night, watermark off, seed 42
Generate an image using seedream-4-5, guidance scale 8, with this reference image: [attach image]

TUI / Web:

Open the TUI (openclaw tui) or web UI (openclaw web) and type your prompt directly. Attach reference images via the paperclip icon in the web UI or drag-and-drop in the TUI.


Video Generation (Seedance)

Videos are generated asynchronously. The agent queues the task and delivers the result as a direct video URL (no file upload, no size limits).

Supported models

Model IDProvider IDTypeMax imagesMax videosMax audios
byteplus/seedance-1-5-pro-251215byteplus-seedance15T2V + I2V2 (first + last frame)
byteplus/dreamina-seedance-2-0-260128byteplus-seedance2T2V + I2V933
byteplus/dreamina-seedance-2-0-fast-260128byteplus-seedance2T2V + I2V933
byteplus/seedance-1-0-pro-250528 (default)byteplusT2V + I2V2 (first + last frame)
byteplus/seedance-1-0-pro-t2v-250528byteplusT2V only
byteplus/seedance-1-0-pro-fast-251015byteplusT2V + I2V2 (first + last frame)
byteplus/seedance-1-0-lite-t2v-250428byteplusT2V only
byteplus/seedance-1-0-lite-i2v-250428byteplusI2V only2 (first + last frame)

Set default video model

# Set in OpenClaw config (persisted)
openclaw config set tools.videoGeneration.defaultModel byteplus/seedance-1-5-pro-251215

Standard parameters

These are passed directly to the video_generate tool:

ParameterTypeDefaultDescription
promptstringText description
modelstringseedance-1-5-pro-251215Model ID (prefix byteplus/ optional)
durationSecondsnumberDuration in seconds (4–10)
aspectRatiostring16:9, 9:16, 1:1, 4:3, 3:4, 21:9
resolutionstring720p480p, 720p, 1080p
imagesimage[]Input reference images (for I2V)
imageRolesstring[]Semantic role per image: first_frame, last_frame, reference_image
filenamestringOutput filename

Seedance-specific parameters via providerOptions

Low-level controls are passed through providerOptions and forwarded to the Seedance API as-is:

KeyTypeDefaultDescription
seednumberrandomFixed seed for reproducible results
watermarkbooleanfalseAdd watermark to output
camerafixedbooleanfalseLock camera position (no camera movement)
draftbooleanfalseDraft mode: forces 480p, faster generation
firstFrameImageUrlstringURL of first-frame reference image (1.0 models)
lastFrameImageUrlstringURL of last-frame reference image (1.0 models)

Text-to-video (T2V)

Command line:

BYTEPLUS_API_KEY=your-key openclaw agent \
  --message "Generate a 5 second 16:9 720p video: golden wheat fields swaying in the breeze at sunset"

In conversation (TUI/web):

Generate a 5 second video of ocean waves crashing on a rocky shore, 16:9, 720p
Make a 10 second 1080p timelapse of a city at night with no watermark, seed 42

With providerOptions:

Generate a 5 second video of a spinning galaxy, use providerOptions: {"seed": 123, "camerafixed": true}

Image-to-video (I2V) — Seedance 1.5 Pro

Seedance 1.5 Pro supports at most 2 input images: first_frame and last_frame.

In conversation:

Generate a video starting from [attach image], mark it as first_frame, 5 seconds, 720p
Generate a video transitioning from [image A] to [image B] — first image is first_frame, second is last_frame

Image-to-video and multi-reference — Seedance 2.0

Seedance 2.0 supports up to 9 reference images, 3 reference videos, and 3 reference audios.

In conversation:

Generate a 5 second video using dreamina-seedance-2-0-260128, with these 3 reference images [attach]

Image-to-video (I2V) — Seedance 1.0 models

Seedance 1.0 models support at most 2 input images (first_frame + last_frame) via providerOptions:

Generate a 5 second video using seedance-1-0-pro-250528, providerOptions: {
  "firstFrameImageUrl": "https://example.com/frame.jpg"
}

Draft mode (faster generation)

Generate a quick draft video of a robot walking through a forest, draft mode

Draft mode forces 480p and skips quality refinement passes, reducing wait time significantly.

Usage in TUI

openclaw tui
# → Type your video generation prompt in the message box
# → The agent queues the task; the video URL is delivered as a message when ready

Usage in Web UI

openclaw web
# → Open http://localhost:PORT in a browser
# → Type your prompt; attach reference images via the paperclip icon
# → The generated video URL is returned as a clickable link in the chat

Usage via command line (non-interactive)

# Basic T2V
BYTEPLUS_API_KEY=your-key openclaw agent \
  --session-id vid-$(date +%s) \
  --message "Generate a 5 second video: aurora borealis over a snowy tundra, 16:9 720p"

# With explicit model and providerOptions
BYTEPLUS_API_KEY=your-key openclaw agent \
  --session-id vid-$(date +%s) \
  --message "Generate a 5 second video of daisy fields, model byteplus/seedance-1-0-lite-t2v-250428, aspectRatio 16:9, resolution 720p, providerOptions {\"seed\": 42, \"watermark\": false}"

Note: Video generation is asynchronous and takes 60–300 seconds. In --message (non-interactive) mode, the CLI may exit before the result is delivered. Use the TUI or web UI for reliable end-to-end delivery, or run the gateway in the background with the Mac app.


Requirements

  • OpenClaw >= 2026.3.22
  • A BytePlus API key with ARK access (Southeast Asia region by default; China region requires baseUrl override)

Troubleshooting

SymptomCauseFix
API key not configuredNo key foundSet via openclaw onboard or BYTEPLUS_API_KEY env var
Plugin not loadedGateway not restartedRestart via OpenClaw Mac app
Video generation times outLong-running async taskUse TUI or web UI instead of --message mode
model not found errorWrong model ID formatPrefix model ID with byteplus/ or omit prefix entirely
China region API errorsWrong base URLSet baseUrl to https://ark.cn-beijing.bytedance.com/api/v3

License

MIT

源码与版本

源码仓库

xieyongliang/openclaw-plugin-byteplus-modelark

打开仓库

源码提交

4eec6a9

查看提交

安装命令

openclaw plugins install clawhub:xieyongliang-byteplus-modelark

元数据

  • 包名: xieyongliang-byteplus-modelark
  • 创建时间: 2026/04/03
  • 更新时间: 2026/04/12
  • 执行代码:
  • 源码标签: feat/video-gen-provider

兼容性

  • 构建于 OpenClaw: 2026.4.11
  • 插件 API 范围: >=2026.3.22
  • 标签: latest
  • 文件数: 16