Edit video through chat with subtitles and export
Posted on Mar 18 • Edited on Mar 22. Task: The author built an OpenClaw skill named video-editor-ai to edit videos via chat by wrapping the NemoVideo backend. They implemented a SKILL.md interpretation layer to convert GUI-centric backend responses (e.g., "click Export") into direct API calls. A routing table prevents unnecessary backend credit consumption for simple actions like checking balance or exporting. The agent handles SSE streams, filtering out internal tool calls and heartbeats while translating text responses for the user.; tools: OpenClaw. Human role: The original post does not say how a person stayed involved.; limitations: The backend charges credits for processing, so incorrect routing can waste resources. Anonymous tokens are rate-limited to one per client per 7 days, requiring persistent client ID storage. Aspect ratio changes require full regeneration rather than editing. The source is a secondary blog post; actual production stability and cost metrics are not independently verified..
Tools
Published on Agent Work Atlas:
Overview
Posted on Mar 18 • Edited on Mar 22
Workflow
The author built an OpenClaw skill named video-editor-ai to edit videos via chat by wrapping the NemoVideo backend. They implemented a SKILL.md interpretation layer to convert GUI-centric backend responses (e.g., "click Export") into direct API calls. A routing table prevents unnecessary backend credit consumption for simple actions like checking balance or exporting. The agent handles SSE streams, filtering out internal tool calls and heartbeats while translating text responses for the user.
Reported result
The workflow successfully enables headless video editing, but the author reports that about 30% of edit operations return no text, only internal tool calls. An initial retry logic caused duplicate edits and double credit charges, which was resolved by implementing a state-diffing approach to verify changes. The skill requires specific keyword optimization in the description to appear in vector search results.
Human role
The original post does not say how a person stayed involved.
Limitations
The backend charges credits for processing, so incorrect routing can waste resources. Anonymous tokens are rate-limited to one per client per 7 days, requiring persistent client ID storage. Aspect ratio changes require full regeneration rather than editing. The source is a secondary blog post; actual production stability and cost metrics are not independently verified.
Try it
I want to build an OpenClaw skill that wraps a GUI-based video API for chat interaction. First, analyze the API documentation to identify endpoints for state retrieval, editing, and rendering. Create a SKILL.md file with a routing table that directs simple queries (balance, export) to direct API calls and complex edits to the SSE stream. Implement a state-diffing mechanism: before sending an edit command, fetch the current project state; after the stream closes, fetch the state again and report the difference to the user to handle silent responses. Ensure the skill description includes keywords like 'subtitle', 'export', and 'edit' for search indexing. Run a dry run with a sample video project to verify that 'remove background music' triggers a state change report without duplicate API calls. Confirm with me before deploying to production.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.