Higgsfield Seedance Operator

API & MCP Connect scripts and AI agents to the generation queue

TRUSTED AUTOMATION

REST API

Stage jobs and read their status with an automation token issued by the administrator.

Stage and inspect a video job

export HIGGSFIELD_QUEUE_API_URL="https://videomaker-api.marketmaker.cc"
export HIGGSFIELD_QUEUE_WORKER_TOKEN="ask-an-admin"

curl --fail-with-body --silent --show-error \
  -H "Authorization: Bearer $HIGGSFIELD_QUEUE_WORKER_TOKEN" \
  -F "model=seedance-2.5" \
  -F "prompt=Slow push-in on the subject" \
  -F "duration_seconds=15" \
  -F "resolution=720p" \
  -F "unlimited=true" \
  -F "files=@/absolute/path/start.png" \
  -F "files=@/absolute/path/end.png" \
  "$HIGGSFIELD_QUEUE_API_URL/api/worker/jobs"

export JOB_ID="copy-job-id-from-the-response"
curl --fail-with-body --silent --show-error \
  -H "Authorization: Bearer $HIGGSFIELD_QUEUE_WORKER_TOKEN" \
  "$HIGGSFIELD_QUEUE_API_URL/api/worker/jobs/$JOB_ID"

Seedance files: start frame, end frame, then optional references.
Seedance 2.0: up to 8 seconds, 720p or 1080p.
Seedance 2.5: up to 15 seconds, 720p only.
GPT Image 2: use gpt-image-2 and omit duration and resolution.

Keep the token in local environment or protected client config. Never paste it into prompts or publish it.

LOCAL STDIO SERVER

MCP server

Install the project on the machine that owns the media files, then add this server to an MCP client.

uv sync --all-groups && uv run camoufox fetch

JSON MCP client configuration

{
  "mcpServers": {
    "higgsfield-auto": {
      "command": "/absolute/path/to/higgsfield-auto/.venv/bin/higgsfield-auto-mcp",
      "env": {
        "HIGGSFIELD_AUTO_ROOT": "/absolute/path/to/higgsfield-auto",
        "HIGGSFIELD_QUEUE_API_URL": "https://videomaker-api.marketmaker.cc",
        "HIGGSFIELD_QUEUE_WORKER_TOKEN": "ask-an-admin"
      }
    }
  }
}

stage_generation: Validate and stage a job from absolute local file paths.
generation_status: Read state, verified settings, errors, and result paths.
run_next_generation: Explicitly run one job in visible Camoufox; allow several minutes.

Before the first run, sign in once with uv run higgsfield-auto login. Clients that use TOML need the same command and environment values shown above.

The MCP server runs locally over stdio. Staging does not submit or spend credits; only run_next_generation can click Generate. In central mode it runs Seedance jobs; GPT Image jobs remain for the administrative image worker.

QUEUE / VIDEO + IMAGE

Render pipeline

0 jobs

Active0

Queued0

Waiting0

Done0

Failed0

Current active job Loading queue

LATEST OUTPUTS

Results ready to view

0 files

NEW TASK

Create generation

Model Prompt

Start frameRequired

Choose start imagePNG, JPG or WebP

End frameRequired

Choose end imagePNG, JPG or WebP

Duration

Seedance 2.5 Unlimited supports up to 15 seconds at 720p.

Reference images

Optional for video. Add characters, products, style frames, or any other visual references.

0 added

Add reference imagesSelect many files at once or add them in several batches

Unlimited mode

WATCHDOG / LOCAL

Polling cycle

JOB INDEX

Generation queue

State	Model	Job	Inputs	Request	Created by	Updated	Open
Loading generation jobs

PROCESS TRACE

Live logs

Waiting for log stream 0 B

Connecting to local process log...