Mission
Teach young people to make films with AI, and to make them well.
The tools are here. Fourteen-year-olds can already generate a shot in a minute. What they can't get anywhere is a place that teaches the craft around the button: what a cut is for, whose face is in it, what it cost, and what to do when the machine gives you something you didn't ask for. Imaje is that place. A studio built for cohorts, where the film comes first and the model is a crew member.
Three promises
Students write the brief, draw the board, choose the look, and direct the take. Explore cheaply, compare honestly, finish deliberately. The messy first takes are the lesson, not a mistake.
Every model is rated on what it was trained on, what its licence allows, where it runs, and what it discloses. Every real person has a signed release before they appear. Every take has a receipt.
We don't call Imaje carbon neutral, because we can't measure it yet. We rate what each model discloses, never invent a number, and are building towards measuring every take on machines running on clean power. We don't buy carbon credits.
What integrity means here
- Consent is specific. A character built on a real person needs a release that names the person, the uses, and the lane it permits. Revoke it, and generation stops. Nothing about an under-18 is ever cloned: voice, face, or likeness.
- Thirteen and up. Imaje isn't for children under 13. Whoever invites a student confirms they're 13 or older, and the student confirms it again when they join. Under-18s are protected as minors throughout.
- Other people's work stays theirs. Prompts are also checked for well-known characters, brands, famous people and living artists' styles. Students get a reason and a way to make the idea their own.
- Prompts pass a gate. Before a prompt reaches a model, it's checked against the school's content tier. A refusal costs nothing and comes with one sentence a teacher can show the student.
- Models are rows in a register, not a brand. Training-data disclosure, licence, indemnity, likeness risk, hosting region, energy disclosure. Students see the rating on the tile before they spend a token.
- Stock voices, plainly. The ready-made voices come from an open model trained on public-domain and openly licensed recordings, plus some speech generated by commercial voice models whose sources aren't public. That second part isn't licensed, and we're working on replacing it: next is a set of voices performed by paid actors who agreed to this use, on a model trained only on public-domain speech and their recordings.
- Receipts are permanent. Which model, which version, which prompt, whose consent, what it cost, what it likely used. Kill a take and it leaves the bin, never the record.
Our footprint, honestly
Imaje isn't carbon neutral, and we won't say it is until we can show the numbers. Today every picture and clip is made on a rented service that publishes no energy figures, and almost no video model publishes what a generation costs in energy or water. Making a number up would be a lie, so we don't. Each model carries a disclosure tier: A when the method is published,B when only independent research exists, C when there's nothing. Today nobody is A.
What independent research does show: a short clip from the family of video models we use measured a few watt-hours on one modern GPU, and the energy grows roughly with the square of a clip's length. Doubling a take's length roughly quadruples its energy. So we teach it on day one: plan the cut, then generate.
The path we're on
- Estimate, labelled as an estimate. A range of energy, carbon and water for each take, from published measurements, with the method on a public page.
- Use less. Quick drafts in class by default, short takes, and a final render only for the takes a student keeps.
- Measure. Once rendering runs on our own machines, read the actual energy of every take and show it on the receipt, discarded takes included.
- Clean power at the source. Run those machines where the grid itself is hydro or geothermal, keeping student data in the US. Clean electricity, not certificates: we don't buy offsets or carbon credits to make the numbers look better.
Why open models
The everyday routes are open-weight. That matters for a school in three plain ways. A student's film and the faces in it can stay on infrastructure the school chooses, rather than becoming someone's training data. The model can be inspected and, on our own machines, metered, which is the only honest route to a real energy number. And the same family carries a cut from first sketch to final take, with camera logic, keyframes, extension and synced sound, so the look holds instead of changing every time a student changes tools.
Closed models stay on the menu for a finish that wants them, with their ratings shown, never as the default. Openness here is not a slogan; it's what lets us keep the promises above.