Edit raw footage by describing the video you want
Cardboard goes from raw footage to an edited video from natural-language instructions.
Done with Gemini
What they did
You upload footage, a prompt, or both, and Cardboard builds an editable first cut. You then refine it through chat or on the timeline. The agent can trim intros, cut dead air, filler words and retakes, add captions, reframe to 9:16, and source B-roll. In the site's documentary example, it transcribed a reference film and matched its cuts, using archival, stock and generated shots.
How it went
The sample documentary took 12m 36s of agent work and came out at 65 seconds with 73 shots, 29 of them generated, and 13 narrated lines. Every shot stays a separate clip you can swap.
Worth knowing
Generative AI is excluded from the $5 Lite plan. Voice cloning, web-sourced B-roll and 4K export need Creator at $30 for the first month, then $59 a month.
Try it yourself with Gemini
I have raw footage in [folder or Google Drive link]. Review it and produce an edit plan for a [length]-minute video about [topic and tone]: which clips to keep, the order, cuts, and suggested captions. Finished means a clear shot-by-shot list with timestamps I can follow in [my editing app]. Don't delete or change any original files.
Source
Read the original