The honest framing: AI video models are not long-form tools yet. They generate seconds. A "long AI video" is a sequence of short generations that were planned to fit together — and planning is the part that decides whether it works.
Why long AI videos are hard
Three problems compound as a piece gets longer. Subjects drift, so the face in scene four is not quite the face in scene one. Lighting changes between generations, so cuts feel like different productions. And narrative attention is unforgiving on social, where a weak second scene kills everything after it.
None of these are solved by asking for a longer clip. They are solved by structure.
Plan the scenes before you generate
Write the beats down first — six to eight lines, one per shot:
- Hook — the strongest image you have, first.
- Setup — who this is and where they are.
- Turn — something changes.
- Consequence — what that change does.
- Resolution — the payoff, short.
Each line should say what is in frame, what moves, and how it is lit. If a beat needs a paragraph, split it into two scenes.
Build it in Rigel
Generate the scenes one at a time, in order, and save each one to your camera roll.
For each scene, keep three things fixed and change only the fourth: same source photo, same character description, same lighting family, different action or location. Rigel's multi-scene storytelling is built for exactly this — switching location mid-story while the subject stays the subject.
The underlying generation step is the one described on the photo to video page. If your story needs dialogue, generate the speaking beats with lip-sync as described on the talking photo page.
Keeping a character consistent
- Reuse one source image for every scene featuring that person. This does more than any prompt wording.
- Repeat the description verbatim. "Woman, early thirties, dark curly hair, olive green jacket" in every prompt, every time.
- Keep wardrobe and time of day stable unless the story explicitly changes them.
- Cut on motion. A cut that lands mid-movement hides small inconsistencies that a static cut would expose.
Multi-reference generation in frontier models has made consistency substantially easier than it was a year ago — see the Seedance 2.0 page for what changed. Our longer write-up on the craft side is in Mastering the art of creating long AI videos.
Export for Reels, Shorts and TikTok
Vertical or square, clips saved to your camera roll, assembled in whatever editor you already use. Two rules that matter more than polish: put the strongest scene first, and cut the piece shorter than feels comfortable. Most multi-scene AI stories that underperform are thirty seconds of idea stretched to ninety.
Cost
You pay per generation, so a multi-scene piece costs as many generations as it has scenes. Rigel is free to download on iOS 16.6+ and Android with starter credits included; credit packs and a subscription are available in-app. Every generation shows its credit cost before you confirm — worth watching when a story runs long.