Generated video is the most over-hyped and under-controlled tool in the room right now. Used raw, it gives you that uncanny mush everyone can already spot a mile away. Used well, it is just another source on the timeline, no more precious than a stock clip or a screen grab.
I treat it like a power tool. Useful, fast, and capable of taking your hand off if you respect it too little. The rule in my room is simple: AI generates material, humans make decisions.
Generate ingredients, not dishes
The mistake is asking a model for a finished shot. You get something that looks almost right, which is worse than something that looks wrong. Instead I ask for ingredients: a texture, a transition element, a frame study, a background plate I can grade and degrade until it belongs.
- Generate textures and overlays, then composite by hand
- Use it for frame studies and look tests before a shoot
- Treat every output as raw footage that still needs an edit
Break it on purpose
Generated footage is too clean and too smooth, which is exactly why it reads as fake. So I rough it up. Grain, real lens artefacts, scanlines, a hard grade, the same colour discipline I use everywhere else. By the time a generated plate has been through the finish, it looks like it was shot, not summoned.
The future without the slop. AI as a tool, kept on a leash, finished by hand.
Be honest about it
I tell clients where it is used, every time. Not because anyone demands it, but because trust is the whole business. The goal was never to fake a shoot. It was to give an editor more raw material to make real decisions with, faster.
AI will keep getting better at producing average. The job, same as it ever was, is to be the one in the room with taste, deciding what stays and what gets thrown back up.