A provenance report is a record of exactly how a piece of AI-generated content was made, the model that produced each shot, the prompt and parameters behind it, and the quality score it earned, packaged so you can hand it to a client alongside the final cut. It answers the question every serious buyer eventually asks: not just what did you make, but how did you make it, and can you prove it? In CoreReflex, that report is not something you assemble by hand after the fact, it falls out of the portable trace the platform already keeps on every generation.
What a provenance report is
When the Director generates a film, every shot rides a portable trace, that trace captures the inputs and the outcome of a generation in a structured, reproducible form. A provenance report is simply that trace, surfaced for review, turned from an internal audit record into a client-facing deliverable that travels with the video.
The distinction matters. Most AI tools hand you a finished file and nothing else. The moment a client, a legal reviewer, or a brand-safety team asks how it was produced. You are reconstructing the answer from memory. A provenance report removes the guesswork because the platform recorded the answer at generation time, shot by shot.
What goes into the report
A useful provenance report is concrete. For each shot in the cut, it surfaces the facts that let someone reproduce and trust the result:
| Field | What it tells the client |
|---|---|
| Model | Which Vertex model produced the shot. Veo, Kling, Imagen, Lyria |
| Prompt | The exact text the shot was generated from |
| Parameters | Camera move, duration, seed, aspect ratio, and other settings |
| Score | The quality-gate result, prompt match, sharpness, motion coherence, text legibility, on-brand fit, claims risk |
| Outcome | Whether the shot passed first time or was selectively regenerated, and why |
Because CoreReflex runs the entire stack on Google Vertex AI, the model field is a real, specific answer rather than "a third-party API." That end-to-end ownership is a big part of why an owned AI stack wins client work, the provenance is complete because nothing in the pipeline is rented from an account you cannot see into.
Why clients are asking
The demand for provenance is not academic. It comes from real pressure that lands on the people buying creative:
- Brand safety. A brand team needs to confirm an asset is on-brand and free of claims it cannot back up. The quality gate's on-brand and claims-risk checks are already in the score, the report just makes them visible.
- Legal and compliance. Regulated industries increasingly need to document how marketing material was produced. A reproducible trace is a far better answer than "our agency made it."
- Trust. A client paying for AI-assisted work wants assurance it was made deliberately, not rolled off a slot machine. Showing the plan, the scores, and the fixes demonstrates exactly that.
This is the opposite of the "generate and hope" approach to AI video, where a clip is handed over with no record of how it came to be. A provenance report is the receipt that the work was governed.
From portable trace to client deliverable
The report is a byproduct of the Director loop, not an extra step. Here is where each piece comes from:
- PLAN. Before any credits are spent, the Director boards the shots and the plan itself is evaluated. That plan evaluation happens before generating, so the report's story starts with intent, not just output.
- PRODUCE. Each shot is generated on a specific Vertex model with recorded prompt and parameters. Those are the rows in the report.
- CRITIQUE. The quality gate scores every shot. Passes and selective regenerations are both logged, so the report shows not just the final result but the path to it.
- ASSEMBLE. The deterministic render path compiles the approved shots. Because the same manifest renders to the same cut, the report is reproducible, anyone can replay it, this idea is covered in depth in our explainer on render provenance in AI video.
The upshot: you do not write the report, you export it. The facts were captured while the film was being made.
How to share it with a client
A provenance report is most useful attached to the deliverable it describes. Practically. That means handing over the final cut and the trace together, so the client can match scores to shots. You decide the altitude: a summary view for a marketing stakeholder who wants the headline, "every shot passed the on-brand and claims-risk checks", or the full per-shot detail for a legal or compliance reviewer who needs to see the prompts and parameters.
Because the trace is portable, it is not locked inside the editor. It travels with the asset, which is what makes it a deliverable rather than a dashboard you have to grant access to. Pair it with the finished video in whatever format the client needs, and if they are publishing across surfaces, the same cut can be exported for vertical 9:16 on Reels and Shorts without changing the provenance behind it. The full detail of what the trace captures lives in the developer docs.
What a provenance report is not
Honesty matters here. A provenance report documents how your content was generated inside CoreReflex, the models, prompts, parameters, and scores the platform recorded. It is not a watermark-detection service, a legal opinion, or a guarantee about content made elsewhere. Its value is specific and real: it makes your own production process transparent and reproducible, so you can stand behind the work and prove the work was governed by a quality gate rather than left to chance.
Frequently asked questions
What goes in a provenance report?
For each shot in the finished cut, it lists the model that generated it, the exact prompt, the generation parameters (camera move, duration, seed, aspect ratio), and the quality-gate score, covering prompt match, sharpness, motion coherence, text legibility, on-brand fit, and claims risk. It also records whether a shot passed first time or was selectively regenerated and why, so the report shows the path to the final result, not just the result.
Can I share how an AI video was made with clients?
Yes. Because every generation rides a portable trace, youcan surface that trace as a client-facing report and hand it over with the final cut, you choose the level of detail, a summary for a marketing stakeholder or full per-shot data for a compliance reviewer.
Is the report reproducible?
Yes. The deterministic render path means the same manifest produces the same cut, and the trace records the model, prompt, and parameters for each shot. That combination is what makes the report replayable rather than a static after-the-fact summary.
Does the report prove a video is safe to publish?
It documents that each shot passed the quality gate's on-brand and claims-risk checks, which is strong evidence the work was governed. It is not a legal opinion or a substitute for your own review, but it gives reviewers a concrete, auditable record to evaluate.
Hand over the receipts, not just the file
Client work runs on trust, and trust is easier to earn when you can show your work. A provenance report turns CoreReflex's portable trace into something you can put in front of a client, the models, the prompts, the parameters, and the scores behind every shot, reproducible on demand. Browse the rest of our how it works guides to see how the Director loop produces that record by default.
Generate a film and see the trace for yourself. It is free to start, no credit card. Try CoreReflex.