tongflow-api-bytedance

June 28, 2026 · View on GitHub

Official TongFlow plugin. Video generation via Volcengine ArkDoubao Seedance 2.0 (ByteDance).

Capabilities

Implements these ABI slots (runs locally as a Python process, no GPU; calls the Ark API):

  • Video generation (text-gen-video) — video from a text prompt.
  • Image-to-video (image-gen-video) — animate a still image (first frame) into motion.
  • First/last-frame video (image-image-gen-video) — two key images interpolated into a clip.
  • Image + audio → video (audio-image-gen-video) — multimodal reference: image + audio → video.
  • Images → video (images-gen-video) — free multi-image reference fusion: up to 9 reference images + text → a new video.

Output includes audio by default (Seedance generates audio + video jointly).

Credentials

Add in TongFlow Settings (gear icon, top-right):

KeyRequiredNotes
ARK_API_KEYVolcengine Ark API key. Create one in the Ark console. Requires an enabled Seedance 2.0 model (account balance ≥ ¥200 or a resource pack).
SEEDANCE_MODELoptionalModel id. Default doubao-seedance-2-0-mini-260615. Other tiers: doubao-seedance-2-0-fast-260128, doubao-seedance-2-0-260128 (only the full model supports 1080p/4K).
SEEDANCE_RESOLUTIONoptional480p / 720p (Mini/Fast) or up to 4k (full). Default 720p.
SEEDANCE_RATIOoptionalFallback aspect ratio when a node has no explicit width/height (e.g. 16:9). Default adaptive.
SEEDANCE_GENERATE_AUDIOoptionaltrue/false. Default true.
SEEDANCE_WATERMARKoptionaltrue/false. Default false.
SEEDANCE_BASE_URLoptionalOverride the Ark endpoint. Default https://ark.cn-beijing.volces.com/api/v3.
SEEDANCE_POLL_TIMEOUT_SoptionalMax seconds to wait for a generation. Default 600.

Values are stored locally and take effect without a restart.

Limitations

  • No real human faces. Seedance rejects input images/audio containing real people's faces (content-safety审核). Use model-generated or non-photoreal assets.
  • Mini / Fast cap at 720p. Use SEEDANCE_MODEL=doubao-seedance-2-0-260128 for 1080p/4K.
  • Audio ≤ 15s for audio-image-gen-video.
  • Up to 9 reference images for images-gen-video.
  • Video-input slots are not supported yet (video-edit, video-image-gen-video-*, speech-video-gen-video). The Ark API only accepts public URLs / asset IDs for video inputs (not base64), so these require object-storage upload — planned for a later release.

How it works

Each slot creates an async Ark task (POST /contents/generations/tasks), polls (GET /contents/generations/tasks/{id}) until succeeded, then downloads the result mp4 and returns it as an ABI Asset. Image/audio inputs are passed inline as base64 data URLs.