Ninety minutes of generated footage is reachable. Ninety minutes in which one character stays one character is where every current pipeline struggles.
Three different things people mean by this
The phrase covers three very different ambitions, and they have very different feasibility. Being clear about which one you are after saves a lot of wasted effort:
- A feature produced end to end by a model. Not currently achievable at a quality anyone would sit through. Consistency degrades long before ninety minutes.
- A short film made largely with generated shots. Achievable now, and being done. The runtime is short enough that drift can be managed by hand.
- A serialized story told in generated episodes. Achievable now, and the only one of the three that builds an audience rather than a single artefact.
It is worth being direct, because a lot of marketing in this space is not: Tosheo does not produce feature films.
Why long-form is genuinely hard
Generated shots drift. Faces shift between takes, costumes change detail, rooms rearrange themselves, and voices move in pitch and accent. Across a two-minute clip you may not notice. Across ninety minutes the audience notices constantly, and drift becomes the thing they are watching instead of the story.
Long-form also concentrates risk in a way episodes do not. A single change late in a feature can invalidate a great deal of finished work, because there is no natural boundary to contain the change. Move a scene and every subsequent reference to it is suspect.
There is a third problem that gets less attention: review. Nobody can hold ninety minutes of continuity in their head well enough to catch a costume error in scene forty that contradicts scene six. An episode is short enough to actually review.
Why episodes are the achievable path
- Consistency must hold unbroken across the full runtime.
- No natural boundary to contain a late change.
- Too long to review continuity reliably.
- All of the audience risk lands on one release.
- Nothing is publishable until everything is finished.
- Consistency is enforced at each episode boundary.
- A change is contained by the episode it lands in.
- Short enough that a person can actually check it.
- Risk spread across many releases, with feedback between them.
- Episode one publishes while episode four is still being made.
A season also reaches feature length in total. Twenty short episodes is a comparable body of work to a film, produced in units small enough to fix.
What Tosheo does instead
Tosheo produces serialized microdrama seasons: an approved recurring cast and world, a planned season arc, episodes developed through script and storyboard, continuity checks before generation, cost approval before rendering, and shot-level correction after.
The practical route today is to build the world and cast as a season, prove the story with an audience, and treat long-form as a later decision rather than a starting point. That order also happens to be how most film projects get financed, so it is not a consolation prize.
If you are set on long-form anyway
This is not the tool for it, and it would be more useful to say what actually helps. If you are attempting generated long-form elsewhere, the constraints that decide whether it works are:
- Lock your cast as fixed references before any scene is produced, not after.
- Keep locations few and reuse them, because every new place is a new chance to drift.
- Write for close framing. Wide shots with multiple characters are where consistency fails first.
- Break the film into scene-sized units you can regenerate independently.
- Keep a written continuity record. You will not remember scene six by scene forty.
- Budget for correction passes, not just for generation.
Every item on that list is a thing a production system should do for you. That is the argument for treating this as a production problem rather than a prompting problem.
Questions creators ask
Can Tosheo generate a full movie?
No. Tosheo produces serialized microdrama seasons made of short episodes. Feature-length generation is not what the platform does, and consistency over feature length is not a solved problem in any current pipeline.
What is the difference between a movie and a season?
A movie is one continuous piece that has to hold consistency across its whole runtime, with no boundary to contain a late change. A season reaches similar total length in short episodes, each small enough to review and correct, with continuity carried deliberately between them.
Why do AI generated videos look inconsistent?
Because each generation is a fresh attempt unless something outside the model enforces the same references. Faces, costumes, locations, and voices drift between takes without approved references and tracked story state.
Could a season be edited together into a film later?
Episodes are exported as finished cuts, so they can be assembled. Whether the result works as a film is a storytelling question: material written to end on hooks does not automatically read as a feature.
Is a short film a better starting point than a season?
A short film is achievable and can be a good calling card, but it does not build a returning audience. A season is designed to make the next episode wanted, which is the compounding asset.