← 返回 Skills 市场
bowen31337

Ai Media

作者 bowen31337 · GitHub ↗ · v1.0.1
cross-platform ⚠ suspicious
789
总下载
0
收藏
4
当前安装
1
版本数
在 OpenClaw 中安装
/install ai-media
功能描述
Generate photorealistic images, videos, talking heads, and natural TTS audio using GPU-accelerated AI models and scripts on a remote server.
使用说明 (SKILL.md)

ai-media - AI Media Generation

Full-stack AI media generation powered by GPU server (RTX 3090/3080/2070S).

Capabilities

  1. Image Generation — Photorealistic images via ComfyUI (z-image, Juggernaut XL)
  2. Video Generation — Video synthesis via ComfyUI (AnimateDiff, LTX-2)
  3. Talking Heads — Animated talking faces via SadTalker
  4. Voice Synthesis — Natural TTS via Voxtral (whisper.cpp)

GPU Server

  • Host: ${GPU_USER}@${GPU_HOST}
  • SSH Key: ~/.ssh/id_ed25519_gpu
  • ComfyUI: /data/ai-stack/comfyui/ComfyUI/ (port 8188)
  • SadTalker: /data/ai-stack/sadtalker/
  • Voxtral: /data/ai-stack/whisper/
  • Output: /data/ai-stack/output/

Usage

Generate Image

./scripts/image.sh "lady on beach at sunset" realistic
./scripts/image.sh "cyberpunk cityscape" artistic

Arguments:

  • $1: Prompt text
  • $2: Style (realistic|artistic) — optional, default: realistic

Output: Path to generated image (e.g., /data/ai-stack/output/image_001.png)

Generate Video

./scripts/video.sh "waves crashing on shore" animatediff 4
./scripts/video.sh "city traffic timelapse" ltx2 8

Arguments:

  • $1: Prompt text
  • $2: Model (animatediff|ltx2) — optional, default: animatediff
  • $3: Duration in seconds — optional, default: 4

Output: Path to generated video (e.g., /data/ai-stack/output/video_001.mp4)

Generate Talking Head

./scripts/talking-head.sh "Hello, I'm Agent" gentle input.jpg
./scripts/talking-head.sh "Welcome to the future" neutral photo.png

Arguments:

  • $1: Speech text
  • $2: Voice style (gentle|neutral|energetic) — optional, default: gentle
  • $3: Avatar image path — optional, generates default if not provided

Output: Path to talking head video (e.g., /data/ai-stack/output/talking_001.mp4)

Generate Audio

./scripts/audio.sh "This is a test message" en male
./scripts/audio.sh "Bonjour le monde" fr female

Arguments:

  • $1: Text to speak
  • $2: Language code (en|fr|es|etc) — optional, default: en
  • $3: Voice gender (male|female) — optional, default: male

Output: Path to audio file (e.g., /data/ai-stack/output/audio_001.wav)

Models Available

Image Models

  • z-image — 6B params, S3-DiT, photorealistic (downloading, 43% complete)
  • Juggernaut XL v9 — SDXL-based, versatile (7.1GB, ready)

Video Models

  • AnimateDiff — SD 1.5 motion module (512x512, working ✅)
  • LTX-2 — 19B params, high quality (14GB checkpoint ready, Gemma encoder ready)

Talking Head Models

  • SadTalker — Audio-driven head animation (working ✅)

Voice Models

  • Voxtral — whisper.cpp-based TTS (installed)

Dependencies

All dependencies are pre-installed on GPU server:

  • ComfyUI with custom nodes (AnimateDiff-Evolved, VideoHelperSuite)
  • SadTalker with face enhancer
  • Voxtral with whisper.cpp
  • FFmpeg for video encoding

Error Handling

Scripts will:

  • Check SSH connectivity before execution
  • Validate GPU server is running
  • Return meaningful error messages
  • Clean up failed generations automatically

Performance

  • Image: ~10-20s for 1024x1024
  • Video (AnimateDiff): ~20-30s for 512x512, 16 frames
  • Video (LTX-2): ~60-90s for 768x512, 4s @ 24fps
  • Talking Head: ~30-40s for 10s video
  • Audio: ~2-5s for 30s speech

Future Enhancements

  • Batch generation support
  • Style transfer capabilities
  • Video upscaling (spatial + temporal)
  • Multi-language voice cloning
  • Real-time preview streaming

Status: Active development Maintainer: Agent GPU Server: ${GPU_USER}@${GPU_HOST}

安全使用建议
This skill runs generation on a remote GPU over SSH. Before installing or running it: (1) Inspect SKILL.md, scripts, and skill.toml for any hardcoded host or key (README and skill.toml include [email protected] and an alternate SSH key name); do not reuse your private keys or credentials without confirming the intended server. (2) Decide whether you want to connect to your own GPU host — set GPU_USER/GPU_HOST and SSH_KEY_NAME explicitly; do not rely on README examples. (3) Be aware the scripts will upload any avatar files you provide (scp) and run remote code under /data/ai-stack on the server; avoid sending sensitive files. (4) If you will not provide your own server, do not run scripts that might attempt to contact the maintainer's host. (5) Consider asking the maintainer to remove hardcoded hosts/keys and to declare required env vars/credentials in the registry metadata. If you want a lower-risk option, run these tools locally on a machine you control or require a vetted remote service with documented auth and privacy practices.
功能分析
Type: OpenClaw Skill Name: ai-media Version: 1.0.1 Multiple shell scripts (`scripts/audio.sh`, `scripts/talking-head.sh`, `scripts/image.sh`, `scripts/video.sh`) directly interpolate user-provided arguments (e.g., `$TEXT`, `$PROMPT`, `$LANG`) into Python scripts executed remotely via SSH. This creates a Remote Code Execution (RCE) vulnerability on the remote GPU server, as an attacker could craft input to break out of Python strings and execute arbitrary shell commands. While `scripts/image.sh` and `scripts/video.sh` (for 'animatediff' model) currently only print the workflow JSON, the RCE vulnerability is present in the design. The use of environment variables for SSH connection details (`SSH_KEY_NAME`, `GPU_USER`, `GPU_HOST`) also presents a risk if not securely controlled by the agent.
能力评估
⚠ Purpose & Capability
The skill claims to run generation on a remote GPU server (expected) and the scripts implement that via SSH/scp and ComfyUI APIs (coherent). However the registry metadata claims no required env vars or credentials while skill files and SKILL.md clearly expect SSH credentials, GPU_HOST/GPU_USER and an SSH key. skill.toml and README hardcode a third-party host ([email protected]) and an SSH key name in places — this mismatch is disproportionate and surprising for a skill that declares no required secrets.
⚠ Instruction Scope
Runtime instructions and scripts perform SSH connections, remote execution, file upload (scp) of user avatar images, creation of directories, and remote Python execution that reads/writes under /data/ai-stack on the remote host. These actions are within the stated media-generation purpose, but the scripts will transmit user-provided files to the remote server and run arbitrary commands there. Also some parts are only partially implemented (placeholders) so the agent may run ad-hoc commands on the remote host during 'automation pending' steps.
✓ Install Mechanism
Instruction-only skill with no install spec; no remote downloads or archive extraction. This is the lowest install-mechanism risk surface.
⚠ Credentials
The registry lists no required env vars/credentials, yet every script expects SSH access: SSH key path (~/.ssh/${SSH_KEY_NAME:-id_ed25519_gpu}), and host/user via GPU_USER/GPU_HOST. skill.toml and README include an explicit external server ([email protected]) and a different SSH key name (~/.ssh/id_ed25519_alexchen), which is inconsistent and could cause users to accidentally connect to the maintainer's host if they follow README examples. Requiring SSH credentials is reasonable for remote GPU usage, but it should be declared clearly and not mixed with hardcoded third-party defaults.
✓ Persistence & Privilege
The skill does not request permanent 'always' inclusion and does not modify other skills or system-wide configuration. It runs commands only when its scripts are invoked.
如何使用
  1. 确保已安装 OpenClaw（本地或 Docker 部署）
  2. 在对话框中输入安装命令：/install ai-media
  3. 安装完成后，直接呼叫该 Skill 的名称或使用 /ai-media 触发
  4. 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.1
Sanitize personal info from examples
元数据
Slug ai-media
版本 1.0.1
许可证 —
累计安装 4
当前安装数 4
历史版本数 1
常见问题

Ai Media 是什么？

Generate photorealistic images, videos, talking heads, and natural TTS audio using GPU-accelerated AI models and scripts on a remote server. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 789 次。

如何安装 Ai Media？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install ai-media」即可一键安装，无需额外配置。

Ai Media 是免费的吗？

是的，Ai Media 完全免费（开源免费），可自由下载、安装和使用。

Ai Media 支持哪些平台？

Ai Media 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Ai Media？

由 bowen31337（@bowen31337）开发并维护，当前版本 v1.0.1。

推荐 Skills
self-improving agent
Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...
⬇ 463283 · by pskoett
Skill Vetter
Security-first skill vetting for AI agents. Use before installing any skill from ClawdHub, GitHub, or other sources. Checks for red flags, permission scope, and suspicious patterns.
⬇ 259410 · by spclaudehome
Polymarket
Query Polymarket prediction markets. Check odds, find trending markets, search events, track price movements.
⬇ 232503 · by joelchance
Self-Improving + Proactive Agent
Self-reflection + Self-criticism + Self-learning + Self-organizing memory. Agent evaluates its own work, catches mistakes, and improves permanently. Use when...
⬇ 200423 · by ivangdavila
Github
Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
⬇ 191115 · by steipete
ontology
Typed knowledge graph for structured agent memory and composable skills. Use when creating/querying entities (Person, Project, Task, Event, Document), linkin...
⬇ 190156 · by oswalpalash

💬 留言讨论