HappyHorse 1.0 — Pro Pack on RunComfy

runcomfy.com · Text-to-video · GitHub

HappyHorse 1.0 — currently #1 on Artificial Analysis Video Arena (Elo 1333 t2v / 1392 i2v) — hosted on the RunComfy Model API. Native 1080p video with in-pass synchronized audio (dialogue, ambient, Foley) and multi-shot character consistency.

npx skills add agentspace-so/runcomfy-skills --skill happyhorse-1-0 -g

When to pick this model (vs siblings)

If the user said "HappyHorse" / "happy horse video" explicitly, route here regardless.

Prerequisites

1. RunComfy CLI — npm i -g @runcomfy/cli
2. RunComfy account — runcomfy login opens a browser device-code flow.
3. CI / containers — set RUNCOMFY_TOKEN=<token> instead of runcomfy login.

Endpoints + input schema

happyhorse/happyhorse-1-0/text-to-video

How to invoke

Default (16:9 1080p 5s):

runcomfy run happyhorse/happyhorse-1-0/text-to-video \
  --input '{"prompt": "<user prompt>"}' \
  --output-dir <absolute/path>

Vertical short (9:16, 8s, no watermark):

runcomfy run happyhorse/happyhorse-1-0/text-to-video \
  --input '{
    "prompt": "<user prompt>",
    "aspect_ratio": "9:16",
    "duration": 8,
    "watermark": false
  }' \
  --output-dir <absolute/path>

Cheaper test pass (720p):

runcomfy run happyhorse/happyhorse-1-0/text-to-video \
  --input '{"prompt": "<user prompt>", "resolution": "720P", "duration": 3}' \
  --output-dir <absolute/path>

The CLI submits, polls every 2s until terminal, then downloads any .runcomfy.net / .runcomfy.com URL from the result into --output-dir. Stdout is the result JSON. Stderr is progress.

Prompting — what actually works

Describe motion over time, not a still. "A woman turns from the window, walks two paces to the desk, picks up the cup, lifts it to her face, takes a sip" beats "a woman drinking coffee".

Camera + shot in plain English. Front-load the shot: "Wide shot. ..." / "Tracking shot. ..." / "Locked tripod, low angle. ..." works as a real directive. Specify lens feel: "35mm anamorphic", "shallow DOF", "crushed shadows".

One visual beat per clip when iterating. Don't pile up "she walks AND the dog runs AND a car passes". Pick the beat, get it sharp, then layer with multi-shot prompts.

Multi-shot consistency — when describing two beats, restate the anchor at each: "Shot 1: tall woman in red wool coat, blue scarf, in a rainy alley. Shot 2: same woman in red coat / blue scarf, now ducking under an awning." HappyHorse holds the look but needs the anchor.

Audio direction — say what you want to hear: "distant temple bells, footsteps on wet pavement, no dialogue" or "warm friendly tone, English".

Anti-patterns:

• Static-frame descriptions (no temporal verbs) → motion will be vague.
• Conflicting style directions → cancels.
• > 2500 char prompts → degrades.
• Aspect ratios outside the 5 supported → 422.

Where it shines

Sample prompts (verified to produce strong results)

From the model page (cinematic scope):

Wide shot. A lone astronaut in dusty orange suit with blue-gray harness
skis across lunar plain, leaving parallel tracks in gray regolith.
Mid-stride, poles planted, pushing in 1/6th gravity with subtle upward
drift. Fine dust haze along ski tracks. Crescent Earth above lunar
horizon, blue-white glow against black sky. Raw sunlight, crushed
shadows, no fill. 8K photorealistic.

Multi-shot consistency:

Shot 1: Medium close-up. A woman in a navy trench coat enters a
rain-slick neon-lit Tokyo alley, looks left, holds up an umbrella.
Shot 2: Same woman in same navy trench, now under the awning of a
ramen shop, shaking water off the umbrella. Warm interior glow, soft
chatter, gentle rain on metal roof in the audio.

Vertical platform-native:

9:16 vertical short. A barista in a black apron pulls a single
espresso shot, steam rising into the morning sun, rich crema slowly
forming. Close-up handheld, shallow DOF, warm cafe ambience and the
hiss of the steam wand.

Limitations

• Duration cap 15s — for longer narratives, segment into multi-shot prompts and stitch.
• Aspect ratios — only the 5 documented values; ultra-wide cinematic gets cropped or rejected.
• Audio is in-pass only — you can't pass external audio to drive lip-sync. For audio-driven lip-sync, use Wan 2.7 (which accepts an audio_url) or Seedance 2.0 Pro.
• No free image-to-video on this template — i2v is supported by HappyHorse via a separate pipeline; the t2v endpoint here is text-only.

Exit codes

The runcomfy CLI uses sysexits-style codes:

Full reference: docs.runcomfy.com/cli/troubleshooting.

How it works

1. The skill invokes runcomfy run happyhorse/happyhorse-1-0/text-to-video with a JSON body matching the schema.
2. The CLI POSTs to https://model-api.runcomfy.net/v1/models/happyhorse/happyhorse-1-0/text-to-video with the user's bearer token.
3. The Model API returns a request_id; the CLI polls GET .../requests/<id>/status every 2 seconds.
4. On terminal status, the CLI fetches GET .../requests/<id>/result and downloads any URL whose host ends with .runcomfy.net or .runcomfy.com into --output-dir. Other URLs are listed but not fetched.
5. Ctrl-C while polling sends POST .../requests/<id>/cancel so you don't get billed for GPU you stopped.

What this skill is not

Not a self-hosted video runner. Not a capability grant — depends on a working RunComfy account.

Security & Privacy

• Token storage: runcomfy login writes the API token to ~/.config/runcomfy/token.json with mode 0600 (owner-only read/write). Set RUNCOMFY_TOKEN env var to bypass the file entirely in CI / containers.
• Input boundary: the user prompt is passed as a JSON string to the CLI via --input. The CLI does NOT shell-expand the prompt; it transmits the JSON body directly to the Model API over HTTPS. No shell injection surface from prompt content.
• Third-party content: image / mask / video URLs you pass are fetched by the RunComfy model server, not by the CLI on your machine. Treat external URLs as untrusted; image-based prompt injection is a known risk for any image-edit / video-edit model.
• Outbound endpoints: only model-api.runcomfy.net (request submission) and .runcomfy.net / .runcomfy.com (download whitelist for generated outputs). No telemetry, no callbacks.
• Generated-file size cap: the CLI aborts any single download > 2 GiB to prevent disk-fill from a malicious or runaway model output.