AI Video Studio
One workspace over three frontier video models (Sora, Veo and LTX) with a shared asset library tying them together.
The question I was chasing
Could one workspace tame three frontier video models that each behave completely differently?
Sora, Veo and LTX don't agree on much: different strengths, different inputs, different ways of failing. The interesting question wasn't 'can I call a video API', it was whether you could build one coherent place to work across all three instead of juggling three tabs and three mental models.
Why it exists
Generative video is moving too fast to commit to one model, and each provider is good at different things. Working across them by hand means re-uploading references, losing track of what you generated where, and starting from scratch each time. I wanted a studio that held the whole workflow in one place.
The constraints
Three frontier video APIs, each with its own endpoints, parameters and quirks, surfaced in a single UI that doesn't flatten them into a useless lowest common denominator. Generated assets had to be reusable across projects rather than trapped in one render. And the whole thing had to stay coherent as models and endpoints shift underneath it. This is a moving target by design.
The decisions that mattered
An asset library as the connective tissue: static images generated with Nano Banana live there and can be pulled in as reference images across any project, so work compounds instead of resetting.
Support every endpoint in a custom UI rather than hiding behind presets, because the point of a studio is control.
Treat the three video models as interchangeable backends for a render rather than three separate apps bolted together. The workflow is the product, the model is a choice within it.
What it is
A custom workspace and pipeline for generative video. An asset library handles static images generated with Nano Banana, reusable as reference images across projects, and videos render through Sora, Veo and LTX, with every endpoint surfaced in a custom UI. There's a walkthrough above.
Where it landed
A working studio, and my standing answer to 'which video model should I use': all of them, in one place, picked per shot. As the models keep changing, the bet that's held up is making the workflow the product and the model a swappable choice. The next thing I'd build is automatic routing: sending a shot to whichever model handles it best, instead of choosing by hand.
Part of the Rolling Waves work archive.