HappyHorse GTtoken

Purpose

Use this skill to generate text-to-video clips through the user's GTtoken/OpenAI-compatible relay instead of RunComfy, Volcengine, or Aliyun Bailian.

Default target:

• Base URL: https://api.modagent-homing.com/v1
• Model: happyhorse-1.0-t2v
• Endpoint style: OpenAI-compatible chat completions unless the user's relay documentation says otherwise.

Required environment variables

Before running the helper script, confirm one API key environment variable is available:

export GTTOKEN_API_KEY="sk-..."

Optional overrides:

export GTTOKEN_BASE_URL="https://api.modagent-homing.com/v1"
export HAPPYHORSE_MODEL="happyhorse-1.0-t2v"

Accept HAPPYHORSE_API_KEY as a fallback if GTTOKEN_API_KEY is not set.

Workflow

1. Keep OpenClaw's normal Models page on a working chat model, such as deepseek-v4-flash.
2. Do not register happyhorse-1.0-t2v as the default chat model.
3. For video requests, call scripts/generate_happyhorse_t2v.py with the user's prompt.
4. Return the generated video URL or downloaded output path if the relay returns one.
5. If the relay returns a task id instead of a final URL, show the raw JSON and tell the user that this relay requires a polling endpoint from its API documentation.

Quick command

python scripts/generate_happyhorse_t2v.py \
  --prompt "赛博朋克东京夜景，下雨，霓虹灯，镜头缓慢推进，电影感，5秒" \
  --duration 5 \
  --aspect-ratio 16:9 \
  --resolution 1080P \
  --output-dir /tmp/happyhorse

Prompt rules

For best results, ask the user for or infer:

• subject and scene
• motion over time
• camera movement
• duration, usually 3-15 seconds
• aspect ratio: 16:9, 9:16, 1:1, 4:3, or 3:4
• whether sound should be present

Prefer a single clear visual beat per clip. Avoid conflicting styles.

Troubleshooting

• 401 or invalid api key: the GTtoken key is wrong, expired, or not authorized for HappyHorse.
• model not found: the exact model id is wrong or unavailable to this key. Try happyhorse-1.0-t2v from the GTtoken model page.
• HTML / response: the base URL is wrong; use the API base URL, normally https://api.modagent-homing.com/v1, not /console/token or the website homepage.
• Chat model test fails in OpenClaw Models: expected for video models. Use this skill instead of Models for HappyHorse.
• Task id but no URL: the relay is asynchronous and needs a status/result endpoint. Ask the user for the GTtoken HappyHorse API example page.