The most expensive review note is “something feels wrong.” It does not identify a time, a failed requirement, or the next action.
A professional AI video review turns taste into observable checks. Watch the clip in several focused passes, record the first failure with a time code, and decide whether the problem belongs in a rerun, an edit, or the prompt for the next version.
This checklist applies to Wan 3.0 text, image, and reference workflows. It is especially useful for longer clips, where a strong opening can hide drift, factual errors, or audio problems later in the timeline.
Prepare the review package
Keep these items together before anyone comments:
- the exact prompt version;
- source images, videos, audio, documents, or links;
- model and workflow mode;
- duration, aspect ratio, resolution, and audio setting;
- the generated file in its original form;
- intended channel and safe-area requirements;
- the scene promise: what the viewer should understand or feel.
Without that package, a reviewer can judge the clip but cannot diagnose it.
Pass 1: watch without stopping
Do not begin frame-by-frame. Watch once at normal speed with sound, then once muted.
Write only three notes:
- What happened?
- Where did attention drop or confusion begin?
- What image or sound remained after the clip ended?
If you cannot summarize the visible story in one sentence, the problem may be structural rather than cosmetic. A technically clean clip can still fail because it has no readable progression.
Pass 2: story and timing
Review whether the clip has a beginning, development, and landing appropriate to its length.
Check:
- the opening establishes the subject and situation quickly;
- the central action starts early enough to finish;
- every cut or camera move advances the same idea;
- pauses feel intentional rather than empty;
- the final state is visible long enough to read;
- dialogue, effects, and music arrive at the correct beat.
For a time-coded prompt, compare the requested and actual beats. Do not demand frame-perfect timing, but note any shift that changes meaning—for example, a product benefit appearing after the voice has already described it.
Pass 3: identity and continuity
Inspect the subject at the opening, at maximum motion, after any occlusion, and in the last clear frame.
People and characters
- face proportions, age, hair, and defining features;
- clothing construction, color, accessories, and logos;
- body scale and limb continuity;
- gaze direction and emotional progression;
- identity after profile turns, hand contact, or crossing behind objects.
Products and props
- silhouette and dimensions;
- material, seams, controls, openings, and connections;
- logo shape and placement;
- hand-to-object contact;
- configuration before and after movement.
Spaces
- doors, windows, furniture, horizon, and structural lines;
- object positions before and after camera movement;
- foreground and background relationships;
- reflections and shadows consistent with visible sources.
Record the first frame where continuity breaks. Later failures may be consequences of that initial change.
Pass 4: motion and physical contact
Watch at half speed when necessary.
Check:
- acceleration and deceleration feel connected;
- feet, wheels, tools, and objects make stable contact;
- weight shifts before a body changes direction;
- fabric, hair, smoke, water, and loose objects respond to motion consistently;
- hands maintain plausible grip and finger count for the shot’s visibility;
- objects do not merge, pass through each other, or change material at contact;
- background motion does not distract from the primary action.
Decide whether the issue affects comprehension. A one-frame peripheral artifact may be editable. A hand changing the product’s shape during the key demonstration usually requires a new generation or a different shot design.
Pass 5: camera and composition
Separate subject motion from camera motion.
Ask:
- does the camera perform the requested move?
- is the subject still readable during that move?
- does camera height, direction, or focal character change unexpectedly?
- do vertical and horizontal lines remain plausible?
- does reframing protect faces, products, captions, and safe areas?
- is the final composition useful for the intended edit?
If the camera obscures a correct action, revise the camera instruction first. Do not rewrite the subject.
Pass 6: audio, voice, and synchronization
Use headphones once and ordinary speakers once.
Dialogue
- words match the approved script;
- language, pronunciation, pace, tone, and speaker are appropriate;
- lip movement and audible phrasing align closely enough for the use case;
- room sound does not change unnaturally between phrases.
Effects and ambience
- visible events have sounds at plausible times;
- off-screen effects do not imply actions that never occur;
- ambience fits the room, distance, and camera perspective;
- transitions do not produce clicks, sudden noise floors, or impossible reverberation.
Music
- entrance and exit support the story;
- music does not mask required dialogue or effects;
- the emotional tone matches the final edit, not merely the opening frame;
- usage rights are documented for any supplied source.
If the visual is approved but audio fails, consider replacing or rebuilding the soundtrack in post rather than sacrificing a strong picture.
Pass 7: text, interfaces, and factual accuracy
Pause on every visible word, number, chart, label, logo, and interface.
Check:
- spelling, punctuation, dates, prices, units, and names;
- chart values and visual relationships;
- brand capitalization, colors, and mark geometry;
- interface controls that appear functional and consistent;
- narration against the approved source document or page;
- claims against the current product, legal, or technical source.
Generated information should be treated as a draft. If text or data is important, replace it with controlled graphics in post or verify it character by character.
Pass 8: rights, consent, and safety
Before publication, confirm:
- you have the required rights to every uploaded image, video, audio file, document, logo, character, and location;
- identifiable people have the necessary consent for the intended use;
- the video does not misrepresent a real person’s words or actions;
- music, voices, brand assets, and source footage are cleared for the delivery channel;
- sensitive, confidential, private, or personal information is absent;
- required disclosure, review, or approval has been completed.
Technical quality does not override publishing responsibility.
Pass 9: delivery check
Review the exported file in the actual context where it will appear.
- aspect ratio and orientation are correct;
- captions and key subjects remain inside platform safe areas;
- the first frame works as a feed or page entry point;
- the last frame does not cut too quickly;
- compression preserves faces, gradients, text, and fine product detail;
- audio level is appropriate beside neighboring content;
- thumbnail, title, alt text, and accompanying copy describe the video accurately.
Do not approve only from the generation preview. Delivery compression and cropping can create new failures.
Classify every note
Use four categories:
| Classification | Meaning | Action |
|---|---|---|
| Accept | The variation does not harm meaning, identity, brand, or delivery | Keep and document |
| Edit | The core generation works; the issue can be corrected safely in post | Crop, retime, replace audio/text, mask, or grade |
| Rerun | The failed element is structural to the generated motion or identity | Revise one prompt/input layer and generate again |
| Reject | Rights, safety, factual, or brand failure makes the asset unusable | Do not publish; resolve the source problem first |
This prevents endless reruns for minor defects and prevents cosmetic edits from hiding a fundamental failure.
Write actionable notes
Use this format:
Time code — requirement — observation — action
Examples:
00:07.4 — Product geometry — the side button doubles when the hand closes — rerun with a fixed camera and simpler grip.00:12.1 — Dialogue — final word begins after the speaker’s mouth closes — rebuild the voice track or shorten the line.00:18.6 — Station layout — clock moves from the central arch to frame left after the orbit — reduce orbit and name the clock position as a location constraint.00:24.0 — Ending — subject exits before the logo card has a full second — retime the edit if clean frames exist; otherwise extend the final hold in the next prompt.
Avoid notes such as “more cinematic,” “make it pop,” or “fix the AI look.” They do not identify the failed requirement.
Final publication checklist
- Story is understandable at normal speed.
- The central action completes and the ending holds.
- Identity, product, wardrobe, and location remain continuous.
- Motion and contact are plausible at the level required by the use case.
- Camera movement protects the main action.
- Dialogue, effects, ambience, and music are synchronized and balanced.
- Every visible word, number, chart, interface, and claim is verified.
- Rights, consent, brand, safety, and disclosure checks are complete.
- The delivered crop, compression, captions, thumbnail, and audio have been reviewed in context.
- Every remaining issue is explicitly classified as accepted or scheduled for correction.
Keep this checklist beside My Creations when reviewing outputs. If the failure begins in the brief rather than the render, return to the Wan 3.0 prompt guide and revise one production decision at a time.