Skill turns one photo and a script into a talking-head video with an AI presenter
An agent skill and script pipeline makes lip-synced talking-head videos of a fixed AI presenter from one photo and a script, then edits them into finished MP4s.
The skill takes a photo of a presenter and a script, then uses Renoise MCP to generate lip-synced talking clips with Grok Video 1.5, according to the README. Scripts are split into segments of 15 seconds or less. Before any credits are spent, the agent quotes the cost and waits for approval, then generates the first segment for the user to check. The finished edit is assembled in Remotion, using one of five editing styles, 48 packaging components, and a cover spliced in as the first frame. The README's 27-second livestream ad example reportedly cost 482 Renoise credits.
The README says the repository was built with Claude Code, and that Claude Code is the only agent tested. Codex, OpenClaw, and WorkBuddy are marked as expected or likely to work, but they are not tested. The README does not say how the work was split between the author and the AI, or how long the project took. The author writes that voice processing cut sibilance spread from 13.2 dB to 5.9 dB in their tests. Grok sometimes burns subtitles into clips, so those segments are re-shot. The text-behind-presenter effect requires macOS and Apple Vision.
In the author's words
built with Claude Code
From the README or the author's post. We checked that the sentence is there, not that it is true.
What to take from it
Put a spending gate in front of any paid generation step: quote the cost, wait for approval, and generate one sample before batching. Splitting a long script into short fixed-length segments also keeps each failure small and easy to re-shoot.
Built with
See it
AI wrote this entry from the project's own README or its author's post. That AI built the project is the author's statement, quoted here. Nobody at IKANGAI ran or tested it. How this list is made