AI Asset Review: Catch Wrong Files and Draft Feedback Before Approval Lock
How Zlaip's AI review assistant scans video, PDF, images, websites, and pasted text against your deliverable and Definition of Done — then turns findings into draft comments you publish when ready.
Part of Approval Lock™ guides guides →
- AI asset review
- review workspace
- definition of done
- client approval workflow
- creative feedback
- wrong file upload
You published v2 of the hero reel. The client opens Review Workspace — and the first comment is "wrong file." Twenty minutes lost, revision round burned, trust dented.
AI Asset Review in Zlaip runs when a review session opens (or on demand). It compares what was uploaded to what the agreement says should be there, flags objective issues on the asset, and suggests draft comments anchored to frames, pages, or text selections. Nothing publishes without a human. Approval Lock™ still requires explicit sign-off.
This guide explains what the scan does, what it does not do, and how it fits revision boundaries and payment gates.
The problem AI asset review solves
Manual review is thorough — and slow. Common failure modes:
- Wrong deliverable uploaded — logo pack instead of social cut, v1 PDF instead of v2
- Objective defects missed — placeholder copy, truncated headlines, obvious contrast failures
- DoD gaps discovered late — approver locks approval, then realizes a checklist item was never addressed
- Scope drift buried in comments — "while you're in there, can we also…" without a change order
Chat and email do not read the file. Generic AI chat does not know your milestone label, scope clauses, or revision limit. Zlaip's scan is agreement-aware.
What gets scanned
When a creative opens a review session, Zlaip queues an asset scan for supported sources:
| Source | AI scan |
|---|---|
| Images, PDF, video, audio (hosted upload) | Yes |
| Website (with snapshot) | Yes |
| Pasted text (blog copy, scripts, decks) | Yes — plain body + extracted text |
| Linked cloud video (Drive, Dropbox, Vimeo OAuth) | Not yet |
| Website URL only (no snapshot) | Scan unavailable until snapshot exists |
Large video assets may take a few minutes; you can keep reviewing while the scan runs in the background.
Two layers: deliverable match + quality findings
1. Deliverable match (wrong-file guard)
First, the model asks: does this file plausibly match this deliverable?
- Full match — proceed with quality review
- Partial match — related but possibly wrong variation; conservative findings only
- No match — clearly wrong subject or format; quality findings suppressed
Creatives also get an early deliverable check when publishing — a warning before the client ever sees the session.
2. Asset findings (actionable issues)
When the asset matches, the scan looks for objective, visible problems:
- Placeholder or lorem text
- Truncated copy or missing sections
- Poor contrast or readability (when accessibility is in scope)
- Obvious completeness gaps vs. Definition of Done items
- Watermarks, resolution, or layout issues you can see on the asset
Each finding includes severity, a short summary, optional evidence quote, and a suggested draft comment with an anchor hint (timecode, PDF page region, text range, pin on image).
Findings are suggestions — dismiss as "not relevant" or "incorrect," accept as a draft, or bulk-accept and edit before publishing.
Human in the loop — always
AI Asset Review is an assistant, not an approver.
- Findings land as draft comments until a reviewer publishes them
- Approval Lock still requires the designated approver and server-side DoD checks
- When revision rounds are exhausted, the scan stops suggesting new revision feedback — blockers and approve-or-not issues only
- Scope drift on published comments is surfaced separately; starting a Change Order™ remains a human decision
The review brief panel may suggest approve vs. request revision with confidence — with a disclaimer. It clusters themes and maps feedback to DoD coverage ("likely met," "needs verification," "possibly missing"). You can add DoD gap drafts the same way as asset findings.
How to use it in a milestone
- Creative publishes the deliverable to Review Workspace (file, website + snapshot, or pasted text)
- AI scan runs — check the AI tab in the sidebar for findings and deliverable match
- Client reviews — pin their own comments; optionally adopt AI drafts after editing
- Generate review brief before decision — themes, DoD gaps, open questions
- Request revision or approve through the normal workflow — revision counter and scope drift still apply
- Approval Lock when DoD is complete and the approver is ready
Compare versions side-by-side when v1 and v2 are the same media type; AI version comparison highlights what changed between sessions.
AI asset review vs. generic vision tools
Tools that "describe this image" do not know:
- Your deliverable label and milestone title
- Scope clauses from Agreement Lock™
- Revision Boundary™ — rounds used vs. included
- That a comment must become an anchored record on the agreement timeline
Zlaip ties scan output to the same session clients already use for pinned feedback — then routes decisions through Approval Lock and Payment Release Gate™.
For format-specific review UX (timecoded video, PDF regions, website pins), see Review Workspace: Video, PDF, and Website.
Setup and limits
Asset scan requires server configuration:
- OPENAI_API_KEY for vision and text analysis
- Object storage for hosted files
- FFMPEG for video keyframe extraction
- Website snapshot upload for staging URL review
Accounts have a daily AI review quota; rescans and brief generation count toward usage. Cached scans reuse results when the same file and context fingerprint has not changed.
Beta note
Review Workspace and AI Asset Review are rolling out in beta. Zlaip is free during beta — no paid tier yet. Enable review in projects where your workspace has the feature flag.