← Back to Skills Marketplace
333
Downloads
0
Stars
1
Active Installs
4
Versions
Install in OpenClaw
/install media-generation
Description
Generate images, edit existing images, create short videos, run inpainting/outpainting and object-focused edits, use reference images as provider inputs, bat...
Usage Guidance
This skill appears to implement the advertised media generation and editing capabilities, but it expects (and will read/use) an OpenClaw configuration file and provider API keys/env vars even though the package metadata does not declare them. Before installing: 1) Inspect ~/.openclaw/openclaw.json (or $OPENCLAW_CONFIG) to see which provider keys would be used; do not provide high-privilege credentials unnecessarily. 2) Be aware the scripts will perform outbound HTTP requests and will download whatever URLs the provider returns (including internal or unexpected endpoints). Consider running the skill in a network-restricted sandbox if you want to limit egress. 3) If you plan to use the skill, create a least-privilege provider account (API key limited to media endpoints) and confirm the provider's URLs and behavior. 4) If the registry entry should declare required env vars/config paths, ask the publisher to add them; lack of declared credentials is an inconsistency and increases risk if you install without inspecting your config. If you want, I can list the specific files and lines that read the config or env vars so you can audit them more easily.
Capability Analysis
Type: OpenClaw Skill
Name: media-generation
Version: 2.2.0
The media-generation skill bundle is a legitimate and well-structured set of tools for AI-driven image and video creation and editing. It features robust scripts for API interaction, asynchronous job polling (generate_video.py), and complex image manipulations like inpainting and outpainting (mask_inpaint.py, outpaint_image.py). The bundle follows secure coding practices by using list-based subprocess calls to prevent shell injection and handles sensitive provider credentials via the standard OpenClaw configuration mechanism (media_request_common.py). No indicators of malicious intent, such as data exfiltration, unauthorized remote execution, or harmful prompt injection, were detected across the code or documentation.
Capability Assessment
Purpose & Capability
The skill's stated purpose (image/video generation & edits) matches the included scripts. However the implementation expects an OpenClaw config with providers (baseUrl and apiKey) and honours env vars like OPENCLAW_MEDIA_PROVIDER, OPENCLAW_MEDIA_IMAGE_MODEL, OPENCLAW_MEDIA_EDIT_MODEL, OPENCLAW_MEDIA_VIDEO_MODEL and OPENCLAW_CONFIG (see references/model-capabilities.md and multiple scripts). The registry metadata declares no required env vars or config paths — that's inconsistent: someone using this skill will need provider credentials/config despite the package not advertising that requirement.
Instruction Scope
SKILL.md instructs the agent to run bundled scripts that (a) read local images and masks, (b) read a config file to obtain provider base URL and API key, (c) send requests to arbitrary provider endpoints, (d) poll status endpoints, and (e) download returned media (including arbitrary URLs, data: URLs, HTML snippets or b64_json). The scripts also create temp files (tmp/images, tmp/videos, mask dirs) and may write temporary JSON files. These actions are appropriate for media generation, but the SKILL.md gives broad discretion (prompt pass-through, using provider-returned URLs) which means the skill can fetch arbitrary remote resources and will read the local provider config — the instructions therefore go beyond a purely local helper and involve network I/O and credential usage that are not declared in metadata.
Install Mechanism
There is no install spec (no external downloads or package installs), which reduces risk from supply-chain downloads. However many executable Python scripts are included and will be invoked at runtime; those scripts will be written to disk as part of the skill bundle. The lack of an install step is coherent with the repo layout but does not remove runtime risk because the scripts perform network requests and file I/O.
Credentials
The code expects and uses provider credentials and config but the skill metadata lists no required environment variables or required config paths. The referenced defaults (in documentation and code) include ~/.openclaw/openclaw.json or $OPENCLAW_CONFIG and env vars OPENCLAW_MEDIA_PROVIDER and OPENCLAW_MEDIA_*_MODEL values. Those are effectively required to call real providers (apiKey/baseUrl). Not declaring these credentials is an incoherence and a safety concern: the skill will attempt to read API keys from user config and then use them to make outbound calls and downloads.
Persistence & Privilege
The skill does not request always:true and does not modify other skills or system-wide settings. It writes outputs and temporary masks under local directories (tmp/images, tmp/videos, mask dir) and uses subprocess to call bundled scripts, which is expected behavior for a script-based media helper.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install media-generation - After installation, invoke the skill by name or use
/media-generation - Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.2.0
Refine SKILL.md structure, clean publish artifacts, standardize reference-image helper entrypoint, and validate/package the skill for cleaner distribution.
v2.1.0
Polish reference-media workflow, rename the main reference helper to reference_media.py with backward-compatible wrapper support, add nsfw image and preset video options, expand smoke coverage, and refresh public examples/docs.
v2.0.0
Refocus the skill into a function-first media tool. Remove prompt expansion, style routing, retry/prompt heuristics, and prompt-oriented references. Keep only transport, provider mapping, polling, retrieval, and reference-image compatibility workflows.
v1.0.0
Initial release: image generation, image editing, mask inpainting, outpainting, reference-image consistency workflows, short video generation, object-focused edits, batch media workflows, robust media fetching, portability notes, and bundled smoke tests.
Metadata
Frequently Asked Questions
What is Media Generation?
Generate images, edit existing images, create short videos, run inpainting/outpainting and object-focused edits, use reference images as provider inputs, bat... It is an AI Agent Skill for Claude Code / OpenClaw, with 333 downloads so far.
How do I install Media Generation?
Run "/install media-generation" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Media Generation free?
Yes, Media Generation is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Media Generation support?
Media Generation is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Media Generation?
It is built and maintained by ljz (@lijiazhen0623); the current version is v2.2.0.
More Skills