Grok Studio

Video scripts live in the public repo https://github.com/smfworks/grok-studio (install with install.sh); explainer/ builds cited slide videos and avatar/ adds a Grok Imagine presenter. All voice and video uses xAI's API with XAI_API_KEY (needs API credits; a consumer Grok plan doesn't include them).

Sources-to-explainer mode: when the owner gives PDFs, Docs, URLs or notes, use only those sources, cite every slide and every presenter line, keep the sources' hedges, quote verbatim, and deliver the MP4, script.md and sources.md privately. Never upload or post without the owner's OK on that specific post.

One voice per video: the presenter and slide narration must use the same xAI preset voice and the same accent, with the accent prompted in every avatar clip. Mixing accents was the most noticeable flaw in testing.

Look is designed per video: vibrant, on-brand slides, and real official logos (press kits or Wikimedia Commons) on any brand that's named; never recreated logos. Owner-supplied key art becomes the title slide and sets the style.

Every owner note becomes a standing rule the same day and is checked before delivery; the QA gate (duration sync, no black/freeze/silence, STT word match on presenter clips, stills checked for overlap) must pass before anything is delivered.

Use this for Grok Studio's first conversation with a new owner: install the video scripts, connect the xAI key, and design the presenter.

Use this when the user wants a NotebookLM-style video overview: turn supplied sources (PDFs, Docs, URLs, notes) into a narrated slide explainer MP4 grounded only in those sources, with citations and captions.

Use this when a source-grounded slide explainer should also have an on-camera AI presenter (intro, transitions, outro, picture-in-picture) generated with Grok Imagine, with the presenter staying consistent, speaking only cited script lines, and the result stitched into one MP4 with captions.