SGLang-Diffusion Video Generation

Generate videos using a local SGLang-Diffusion server (Wan2.2, Hunyuan, FastWan, etc.). Use when: user asks to generate, create, or render a video with a loc...

Security Scan

The skill's code, runtime instructions, and requirements are consistent with its stated purpose (generating video via a local SGLang‑Diffusion server) and it does not request disproportionate credentials or install actions.

Purpose & Capability

Name/description match the included script: the skill only requires python3 and talks to a local SGLang‑Diffusion server (default http://127.0.0.1:30000). No unrelated binaries, cloud credentials, or system paths are requested.

Instruction Scope

SKILL.md and the script instruct the agent to POST prompts and (optionally) base64 input images to the configured SGLang server, poll status, download the resulting MP4, and print a MEDIA: line for auto-attachment. This is appropriate for the purpose, but it means any server URL you point it at will receive your prompt and any provided input image — do not point to untrusted remote servers or supply sensitive files as input images.

Install Mechanism

No install spec is provided (instruction-only with one included script). Nothing is downloaded or written to disk by an installer; risk is limited to running the provided Python script.

Credentials

Only an optional SGLANG_DIFFUSION_API_KEY is referenced (used when the server enforces an API key). No unrelated secrets or multiple unrelated environment variables are requested.

Persistence & Privilege

always is false and the skill does not request persistent/privileged platform presence or modify other skills. It only runs when invoked.

Assessment

This skill appears to be what it says: a Python script that talks to a SGLang‑Diffusion server to generate videos. Before installing or running it: (1) ensure the server URL you use is local or trusted (by default it is http://127.0.0.1:30000); (2) protect any API key you set in SGLANG_DIFFUSION_API_KEY or the CLI, and avoid pasting it into public places; (3) do not pass sensitive images via --input-image since the script base64-encodes and uploads them to the configured server; (4) generated files default to /tmp and the script prints a MEDIA: line which some chat integrations may auto-attach—be mindful of where outputs are posted. Overall the skill is internally consistent and proportional to its stated purpose.

Generate videos via a local SGLang-Diffusion server's OpenAI-compatible API.

Video generation is asynchronous and takes several minutes. The script handles submission, polling, and download automatically.

Prerequisites

• SGLang-Diffusion server running a video model (default: http://127.0.0.1:30000)
• Supported models: Wan2.2-T2V, Wan2.2-I2V, FastWan, Hunyuan
• If the server was started with --api-key, set SGLANG_DIFFUSION_API_KEY env var

Generate a video

python3 {baseDir}/scripts/generate_video.py --prompt "a curious raccoon exploring a garden"

Useful flags

python3 {baseDir}/scripts/generate_video.py --prompt "ocean waves at sunset" --size 1280x720
python3 {baseDir}/scripts/generate_video.py --prompt "city timelapse" --negative-prompt "blurry, low quality"
python3 {baseDir}/scripts/generate_video.py --prompt "dancing robot" --steps 50 --guidance-scale 7.5 --seed 42
python3 {baseDir}/scripts/generate_video.py --prompt "flying through clouds" --seconds 8 --fps 24 --out ./my-video.mp4
python3 {baseDir}/scripts/generate_video.py --prompt "flying through clouds" --server http://192.168.1.100:30000 --out ./my-video.mp4
python3 {baseDir}/scripts/generate_video.py --prompt "cat playing" --poll-interval 15 --timeout 1800
python3 {baseDir}/scripts/generate_video.py --prompt "animate this scene" --input-image /tmp/scene.png

API key (optional)

Only needed if the SGLang-Diffusion server was started with --api-key. Set SGLANG_DIFFUSION_API_KEY, or pass --api-key directly:

python3 {baseDir}/scripts/generate_video.py --prompt "hello" --api-key sk-my-key

Or configure in ~/.openclaw/openclaw.json:

{
 skills: {
   "sglang-diffusion-video": {
     env: { SGLANG_DIFFUSION_API_KEY: "sk-my-key" },
   },
 },
}

Notes

• The script prints a MEDIA: line for OpenClaw to auto-attach on supported chat providers.
• Output defaults to timestamped MP4 in /tmp/.
• Video generation typically takes 5-15 minutes depending on GPU and model size.
• Do not read the video back; report the saved path only.