How to Make Videos Without a Camera

Use it when: You need one clip — a hook, an ad cutaway, a single moment — and you need it in one pass.

1. 01Describe one continuous shot: subject, what it does, camera move, lighting, and the look. One shot, not a sequence — text-to-video models cut badly when a prompt asks for two locations.
2. 02Set the aspect ratio before you generate, not after. Vertical output that was framed for 16:9 loses the subject's head.
3. 03Generate two or three takes of the same prompt before you change the prompt. Variance between takes on one prompt is usually wider than the difference between two prompts.
4. 04Keep the take that has the right motion, even if the details are wrong — details can be fixed downstream, motion can't.

Run it here

Use it when: The finished thing is 30 seconds or longer and needs a beginning, a middle and an end.

1. 01Write the premise as one sentence — who, what changes, and how it ends. Scene splitting works from a premise; it does not work from a mood.
2. 02Lock the recurring characters, props and locations as named assets FIRST. Continuity across scenes comes from reused reference assets, not from repeating adjectives in every prompt.
3. 03Generate scene by scene and accept that some scenes need four takes. A 6-scene piece where one scene is wrong reads as broken; the other five carry nothing.
4. 04Stitch, then add the voice and captions over the finished cut rather than baking narration into each clip.

Run it here

Use it when: You can talk but you can't film — a recorded take, a podcast clip, or a script you're happy to read aloud.

1. 01Cut the audio first and treat it as the spine. Everything visual is timed to it, so re-timing later means regenerating clips.
2. 02Break the audio into beats and write one generated visual per beat — literal illustration of the sentence being spoken beats abstract footage on retention.
3. 03Generate the visuals as separate short clips rather than one long one, so a bad beat costs one clip.
4. 04Lay the clips over the audio, then caption. Captions last because the cut length is what decides where lines break.

Run it here