Claude Motion generates animation with code, drawing 10.8M views for an AI video with no video model
Claude Motion entered beta on October 8 and controls every word, shape and timing by writing code, then exports an mp4; the author used it with Picsart layering to make this video.
10.8 MILLION PEOPLE WATCHED AN AI VIDEO LAUNCH WITH NO VIDEO MODEL IN IT claude motion went into beta on october 8, and the announcement has close to 40,000 bookmarks. it makes short animated explainers and exports an mp4. there is no generated footage in it. claude writes code for every word, shape, chart and timing, so when you change one number, that number moves and the rest of the animation stays where it was. that splits AI video into two jobs. diffusion handles what has to look filmed, like faces, light and a wet street. code handles what has to be right, like a price, a chart, a logo or a line of text. a wrong letter in a generated shot costs a reroll, and the reroll brings a sky and a door the client never approved. in code it is one edit, which is how this video and my last two were made. with @Picsart, i keep the two layers apart. the filmed shot comes from one still through image-to-video, and every word on top stays an editable text layer, so the model never has to spell anything. before you prompt anything, ask one thing about every element in the frame. does it have to look real, or does it have to be right?
This post quotes another post: x.com/claudeai/status/2108271552991252810