ElevenLabs Scribe v2 for precise timestamps
Gemini 3.1 did video comprehension.
FLUX2.MAX for keyrame generation, Opus 4.6 requested four at a time and combined them into one 2x2 grid to pick the best one.
Video is Wan 2.6 with segmented audio input, requesting several takes in parallel. Opus 4.6 asks Gemini 3.1 to review them, as well as the splice results.
Scratchpad fragments were overlaid by Opus 4.6.
The whole pipeline was entirely autonomous and completed on the first attempt - I never reviewed intermediate results. The only piece of feedback that I gave was that scratchpad quotes were first virtually unreadable to humans, being gray on gray, and Opus 4.6 adjusted those.
Gemini 3.1 did video comprehension.
FLUX2.MAX for keyrame generation, Opus 4.6 requested four at a time and combined them into one 2x2 grid to pick the best one.
Video is Wan 2.6 with segmented audio input, requesting several takes in parallel. Opus 4.6 asks Gemini 3.1 to review them, as well as the splice results.
Scratchpad fragments were overlaid by Opus 4.6.
The whole pipeline was entirely autonomous and completed on the first attempt - I never reviewed intermediate results. The only piece of feedback that I gave was that scratchpad quotes were first virtually unreadable to humans, being gray on gray, and Opus 4.6 adjusted those.