AI-generated and AI-augmented footage looks finished until you put it on a calibrated monitor next to live action. Then the problems show: tones that drift from shot to shot, grain and texture that don't match, dynamic range that collapses in highlights and shadows, skin tones that shift, and no consistent colour space to build on. We finish AI material to broadcast and cinema standard and make it sit with the rest of your film.
Why AI footage is hard to finish
Most GenAI still hands you 8-bit SDR — compressed and low in bit depth — from models trained on degraded sources that never learned what clean, high-bit-depth material looks like. That inverts the normal pipeline: instead of flowing downhill from 10–14-bit RAW to delivery, quality has to be rebuilt uphill from a lossy 8-bit start. The gap is real — 8-bit holds 256 values per channel, a 32-bit float master holds 4.3 billion. The frontier is starting to shift: Luma's Ray3 now generates natively in 16-bit ACES EXR. But it's the exception, and the harder problems — consistency, and grades that can't survive a regeneration — remain. Dropping a LUT on 8-bit AI footage doesn't hold up; there's no clean starting point for it to sit on.
How we approach it
- Spatial upscaling — AI-based resolution enhancement to 4K.
- Artifact removal — targeted decompression, with no new hallucination.
- Bit-depth expansion — export to 32-bit EXR in linear light.
The result is a grading-ready 32-bit file — and it isn't an empty container. The AI reconstructs the missing data: it calculates intermediate values, draws on learned reality, and averages across frames. From there it's a normal high-end finish.
One thing worth knowing up front: a grade can't be copied across AI generations. Re-generate a shot for a client change and the old grade no longer fits — every generation has different pixels. We plan for that, so iterative AI work doesn't quietly blow up your schedule.
Why it matters
Audiences and QC both notice when a shot doesn't belong. Done right, AI material becomes a usable part of the production rather than the weak link a viewer's eye snags on. Done wrong, it undermines everything around it.
We've worked on this in the open
Colorado presented "Beyond the Prompt: Colour Grading in the Age of AI" at FilmLight / Berlinale 2026, on exactly these problems and how to solve them. This isn't a sideline — it's where finishing is heading.
Bring us your AI material.