Automated AI video review is the step where a generation system grades its own footage before a human ever watches it, scoring each shot against concrete quality checks and reshooting the failures automatically. In CoreReflex this is the CRITIQUE stage, and it is the part most AI video tools skip entirely. They generate, hand you the result, and leave the judging to you. CRITIQUE inverts that: the studio is the first, harshest viewer of every shot, and only footage that clears the bar reaches your timeline.
What the CRITIQUE stage is
CRITIQUE is the third stage of the agentic Director loop, PLAN → PRODUCE → CRITIQUE → ASSEMBLE. After the Director boards your film and generates each shot, CRITIQUE evaluates that footage shot by shot, assigns each one a score against a fixed set of checks, and routes anything that fails back for regeneration. It runs without you. The point is not to replace your judgment but to spend it well: instead of reviewing a hundred raw shots, you review a cut where the obvious failures have already been caught and fixed.
This is what makes the Director agentic rather than just generative. A generator produces. An agent produces, evaluates its own work against a goal, and acts on the result. CRITIQUE is the evaluate-and-act half of that loop, and it is why a director-led pipeline behaves differently from a single prompt. If the loop itself is new to you, the overview of what an agentic AI video director is lays out all four stages.
The checks every shot is scored against
A vague "looks good" score would be useless. CRITIQUE grades each shot on concrete, named checks, so a failure tells you what failed:
- Prompt match. Does the shot actually depict what the board asked for? A beautiful clip of the wrong thing still fails.
- Sharpness. Is the image clean, or soft and artifact-ridden?
- Motion coherence. Does movement hold together across frames, or does the subject warp and flicker?
- On-screen text legibility. If the shot carries text, is it readable and correctly rendered?
- On-brand. Does the shot fit your brand kit and look like it belongs to the rest of the film?
- Claims-risk. Does the shot imply something you can't or shouldn't claim?
Each check is a gate, not a vibe. A shot that nails composition but garbles its on-screen text fails the legibility gate, and the system knows precisely why. That specificity is what lets CRITIQUE do something more useful than flagging. It can fix the right thing.
Selective regeneration: reshoot the shot, not the film
The most important behavior in CRITIQUE is selective regeneration. When a shot fails, the studio regenerates that shot, not the entire film, this sounds obvious, but it is the difference between a tool you can trust at scale and one you can't.
Consider a twelve-shot explainer where shot seven comes back soft. A start-over model throws away eleven good shots to fix one. CRITIQUE keeps the eleven and reshoots only the seventh, re-scoring it until it clears the gate or escalates. The economics matter: generation costs credits, and selective regeneration means you are not paying to re-render footage that already passed. It also protects continuity, because the last frame of each shot anchors the next, reshooting in place preserves the visual through-line that a full restart would scramble.
The result that lands in your timeline is a film where every shot has individually passed the same bar, assembled in order. That handoff from approved shots to a sequenced cut is the job of the ASSEMBLE stage, which takes only footage CRITIQUE has cleared.
Where CRITIQUE sits in the Director loop
It helps to see the stages as a relay, each with a distinct job, the backbone of how the studio works:
| Stage | Job | Costs credits? |
|---|---|---|
| PLAN | Board the film into shots (role, camera, prompt) | No |
| PRODUCE | Generate each shot on the Vertex stack | Yes |
| CRITIQUE | Score every shot, regenerate failures | Only on regeneration |
| ASSEMBLE | Render approved shots into a graded cut | Render only |
Planning and scoring are free; you pay for generation, that pricing shape is deliberate. It means CRITIQUE can be thorough without punishing you for letting it work, since the scoring pass itself doesn't burn credits. You can iterate on the board and lean on the gate as much as you need.
Why automated review changes the economics of AI video
The honest problem with most AI video is variance. Any single generation might be brilliant or broken, and the only way to know has traditionally been to watch all of it. That makes "AI video at scale" a contradiction, the more you generate, the more you have to manually inspect, and review becomes the bottleneck.
Automated review breaks that constraint. When the system grades and reshoots before you watch, your review time scales with finished films, not with raw shots. A team producing fifty localized variants doesn't review fifty timelines of unvetted footage; they review fifty cuts that have each passed the same gate. The quality floor is enforced by the machine, and human attention moves up to where it is actually valuable, creative direction, not defect-hunting.
For agencies, this floor is also a trust instrument. A consistent, machine-enforced quality bar, including the on-brand and claims-risk checks, is exactly what lets teams prove brand safety to clients instead of asking them to take it on faith.
Provenance: every score is replayable
A score you can't inspect is just an opinion. CRITIQUE attaches a portable provenance trace to every generation, the model used, the prompt, the parameters, and the resulting score. That trace makes the whole review auditable and reproducible: you can see why a shot passed, why another was reshot, and what the winning generation actually was.
Because the trace is portable and the render path is deterministic, an approved film isn't a lucky take you can never recreate. The scores, the prompts, and the manifest together mean the same inputs reproduce the same cut, the foundation of treating the manifest as your video's source of truth. Review stops being a one-time gut check and becomes a durable, replayable record.
Frequently asked questions
What happens during the critique stage?
After each shot is generated, CRITIQUE scores it against concrete checks, prompt match, sharpness, motion coherence, on-screen text legibility, on-brand fit, and claims-risk. Any shot that fails is regenerated selectively, re-scored, and only passed forward once it clears the gate. The stage runs automatically, before you review anything.
Does the director review shots without me?
Yes. The whole point of CRITIQUE is that the studio is the first reviewer, catching and fixing failures before they reach your timeline. You still have final say in the editor, but you are reviewing footage that has already passed an automated quality bar rather than raw, unvetted generations.
Does a failed shot mean the whole video regenerates?
No. CRITIQUE uses selective regeneration, it reshoots only the failing shot and leaves the passing ones untouched. That saves credits, preserves the shots you already like, and protects continuity, since each shot's last frame anchors the next.
Can I see why a shot passed or failed?
Yes. Every generation carries a portable provenance trace, model, prompt, parameters, and score, so you can inspect exactly why any shot cleared or was reshot. Combined with the deterministic render path, that makes the entire review auditable and reproducible.
Let the studio review first
The hardest part of AI video isn't generating footage. It is trusting it. CRITIQUE moves that trust from hope to a scored, replayable gate, so the cut you open is one where the obvious failures are already gone. Describe your film, let the Director board and produce it, and watch CRITIQUE do the first review for you. Start free with no credit card and see the quality gate work on your own shots.