01
One record, bid to P/L
The bid, the schedule, every take, every review and every cost are one set of data.
Publishing a bid creates the tasks, dates, crew bookings and AI budgets. Hours and generation land on the shot they were spent on, and the P/L reads them back. Trackers, generators and review tools each hold one slice.
02
AI spend is a budget line
Token budgets come from the bid, per project, artist and shot.
Soft budgets warn; hard limits stop a take before it runs. AI spend sits in the P/L next to labour. Studios can price AI work and hold to the price.
03
Provenance on every image
Every take keeps its model, inputs and prompt. Every source keeps its rights.
It is recorded as the work happens, with nothing typed in by hand, and it can be reported per show. The record gets more valuable with every frame a studio makes.
04
Model-neutral by design
Loft runs image, video, voice and finishing models from several providers, on the shot.
A better model is a new option on the shot, not a migration. Loft gains from model churn instead of being exposed to it, and it isn't tied to any one lab.
05
Built for agents
The Assistant reads the same data producers do, and can act on it.
Ask what's at risk this week and it names the late shots and who has them, then opens the page or makes the change. A REST API and an MCP server let a studio's own tools and agents in too.
06
Built inside real productions
Loft grew out of films directed by Phil Tippett, who advises the company.
Each feature answers a question a working production asked. Mad Goddess is the first show run on Loft from bid to cost, and the team presented its AI work at VIEW Conference 2026.