LOFTai › Your First Hour
0:55 · What You Built & Where Next
Goal: see the shape of what you just did, and know what's worth learning next. Five minutes.
The loop you just learned
Strip away the specifics and the hour was one loop: intent → generate → review → publish, with everything you learned along the way (a primer, a playbook, a reference role) folded back in so the next pass starts further along.
That loop is the whole app. Editorial, Flows, Composer, Dailies, the Assistant — they're different surfaces onto the same four beats, and they all read the same assets, statuses, and lineage. Nothing you learn later replaces this; it extends it.
What Loft recorded while you worked
This is the part that isn't visible in the moment but pays off in month three:
- Every output is an asset item on a real asset — never a loose file with a hash for a name.
- Every generation kept its model, prompt, seed, and inputs, so any result can be reproduced or regenerated.
- Every derived item recorded what it came from — that's lineage, and it's why "how was this made?" is always answerable.
- Your statuses became the pipeline's signal, and your playbook became standing context the Assistant carries forward.
Hour two — pick one
Assemble a cut
Editorial: build a sequence from shots, where clips point at slots rather than frozen files, so the edit re-conforms as shots get new published takes. Start with Workflow: Assemble a Cut.
See the graph
Flows: the node canvas of a file's generation history. Select any generated file and use View generation flow to watch your own hour play back as a graph.
Build a view
Composer: ask the Assistant for a dashboard — "show this week's submissions" — and it builds one you can keep.
Bring the team in
Manage Team on Project Detail, plus Tasks for assignment and scheduling. Notes and annotations become genuinely useful the moment there are two of you.
Watch the spend
Costing breaks your generation spend down by service, asset, model, and month — worth checking after your first real week.
Go local
If you started in the browser, try the other track: point LoftDesktop at a folder and run generation on your own provider keys.
Habits worth keeping
- One output while you're finding the prompt. Variations only once the setup is right.
- Anything you've typed three times goes in a playbook.
- Set status as you go, not in a cleanup pass later — downstream work reads it.
- Change one variable per iteration, so you actually learn the model.
- Say what each reference image is for, by its @-badge.
Tip: Help is contextual: open the help panel anywhere in the app and it selects the topic for the page — and the tab — you're currently on. You rarely need to come looking.
Note: From here, the Workflows section below covers specific jobs end to end, and the topics after it are the full reference for every page.