Pre-Publish Checklist for AI Images and Video

A generator produces in a minute what used to take a day, but it does not know your product, your audience or what would embarrass your brand. A sixth finger on a hand holding a cup, a product name rendered in broken Arabic letters, a logo that changes shape between two shots: small details that audiences spot quickly and that chip away at trust. This checklist turns review into a fixed habit rather than a mood. Run it before every post, in the same order, and record the decision.

The core rule: every output is reviewed by a human before it reaches another human. The best reviewer is someone who did not make the piece, because the maker sees what they intended, not what was actually produced.

How to use this checklist

View it the way the audience will. Open it on a phone at real size and full screen, then on a large screen to inspect detail. Many defects only show up at one of the two sizes.
Compare against the real reference. Put the actual product photo or the brand guide next to the generated piece. Do not rely on memory.
Go through the nine sections in order. Skip what does not apply (no audio in a still image, for example) and write "n/a" rather than leaving it blank.
Classify each finding: blocker, fix before publishing, or note for next time.
Record the decision in the review log (template below): publish, revise, or discard, with the reviewer's name and the date.

Class	Meaning	Examples
Blocker	Not published until fixed or dropped	Product does not match the real one, a real person's face without consent, wrong price or phone number, another brand's logo
Fix before publishing	A clear flaw, but quick to fix	Typo, late captions, music louder than voice, bad framing crop
Note for next time	Does not hurt the message but worth improving	Lighting slightly cooler than usual, a transition slower than we like

1. Product or service fidelity

The most dangerous flaw in a generated image is not ugliness but accidental dishonesty: a product that looks bigger, more premium or a different colour than what the customer will receive.

Shape and proportions match the real product (size relative to a hand or table, number of pieces, packaging shape).
Colours are close to reality, not saturated to the point of misleading.
Labels and packaging text are correct or fully hidden, with no invented letter-like squiggles.
No components or features the product does not actually have (extra toppings, bundled accessories, a sea view from a venue that is not by the sea).
Quantities are right when they are part of the offer ("6 pieces" shows 6 pieces).
If the image is illustrative rather than the actual product, that is clear from context or stated.

2. Faces, hands and bodies

Finger count is right, and hands hold objects in a physically possible way.
Eyes are symmetrical and look in the same direction, teeth are natural, no features melting into the background.
People in the background are not distorted or half-formed.
Clothing and accessories make sense for the context, culture and season.
No face of a real person, known or unknown, generated without written consent, and no deliberate likeness of a public figure.
Representation is respectful: no demeaning stereotypes of a group, profession or region.

3. Continuity (video and series)

In a video assembled from separately generated shots, what changes between shots draws more attention than what changes within them.

The same character wears the same clothes and has the same hair across all shots.
The product does not change shape or colour from one shot to the next.
Light direction and time of day stay constant unless the change is intended.
Nothing appears or vanishes suddenly, and there is no odd "melting" or morphing mid-shot.
Camera movement is logical and does not jump without reason.
For a series of post images: same style, same palette and the same style sentence in every prompt.

4. Arabic text

At the time of writing (September 2026), most generators still stumble when writing Arabic inside an image: disconnected letters, reversed order, missing dots. The most reliable fix is to generate the image without text and add the text in a design editor.

Letters join correctly and words read right to left.
Spelling is correct, including hamzas and taa marbuta; diacritics, if any, are correct or removed.
Numbers, prices, phone numbers and dates match the source exactly.
Punctuation is Arabic and correctly placed (، ؛ ؟), not mirrored.
The font is a brand font and readable on a phone screen.
Text sits away from the frame edges and from areas covered by the app interface (like buttons, account name, caption bar).
Contrast between text and background is enough to read in sunlight.
Video captions are in sync, and each line stays on screen long enough to read.

5. Logos and trademarks

Your logo is the approved version, not distorted or stretched, in the right colours.
The logo was added in the editor, not "drawn" by the generator from memory.
No logos or marks of other companies crept into the background (on cups, clothes, signs, screens).
No badges or seals implying a certification, accreditation or award that does not exist.
No imitation of a competitor's look that could confuse the audience.

6. Audio and lip sync

Names, place names and numbers are pronounced correctly (brand name, city, price).
Tone and dialect suit the audience; Modern Standard Arabic, if used, is free of glaring errors.
Lip movement matches speech for avatars or talking characters, with no noticeable lag.
Music sits clearly below the voiceover (a common starting point: voice at full level, music at roughly a fifth, then adjust by ear).
No hiss, clicks or abrupt cut at the end.
Any cloned voice or avatar of a real person has written, stored consent.
Music is licensed for commercial use under the tool's or library's terms at the time of production.

7. Timing and pacing

The first three seconds say why the viewer should keep watching.
Every shot serves the message; no extra shot kept just because it looks nice.
Total length suits the platform and purpose.
The call to action stays on screen long enough to read or screenshot.
The first frame (cover) is clear, since platforms may show it as a thumbnail.

8. Permissions, rights and disclosure

Written consent for every real face or voice that appears or was used as a reference.
The generator's terms allow commercial use on your current plan (check them; terms change).
Fonts, music and reference images are licensed.
No "illustrative" generated images of real events without a clear label.
Generated content is disclosed where the audience expects it, the platform requires it, or your organisation's policy says so.
No personal data visible by accident (customer names, numbers, real screens).

9. Final export checks

Aspect ratio fits the use: vertical 9:16 for stories and short video, 4:5 or 1:1 for feed posts, horizontal 16:9 for websites and presentations.
Resolution is sufficient, with no visible compression or banding in gradients.
File format and size are within the platform's limits (check that platform's current limits).
Captions are burned in or uploaded as a separate file, as planned.
The file name is clear and carries a version number (for example zaytouna-reel-v3-final.mp4).
Alt text is written if the image goes on a website.
The source version is saved (editable project and prompts used) so it can be reproduced or edited later.

Review log template

Copy it into a shared file or sheet, one line per piece. The log protects you when someone asks "who approved this?" and, over time, shows you where mistakes keep recurring.

Visual asset review log

Asset: [file name and version]
Type: [image / video / series]        Platform: [platform and aspect ratio]
Maker: [name]                         Reviewer: [name]        Date: [..]
Reference used for comparison: [product photo / brand guide / approved copy]

Sections: 1 Product [ ]  2 Faces [ ]  3 Continuity [ ]  4 Arabic text [ ]
          5 Logos [ ]  6 Audio [ ]  7 Timing [ ]  8 Permissions [ ]  9 Export [ ]
          (mark: ✓ fine / ✗ finding / — n/a)

Findings:
- [description + timestamp or location] — [blocker / fix / next time]
- ...

Decision: [publish / revise and re-review / discard]
Note for next time (what to change in the prompt or process): [..]

Illustrative example

Al-Sindiyan Roastery is a fictional small coffee roaster in Tripoli. It produced a 20-second vertical video launching a new blend called "Sabah al-Mina" (Harbour Morning), plus a 4:5 post image. The reviewer is a colleague who took no part in production; she reviewed on her phone first, then on a laptop.

Where	Finding	Section	Class	Action
Video 0:04	The coffee bag is taller and thinner than the real one, and light brown instead of olive green	1	Blocker	Regenerate the shot from a real photo of the bag instead of a text description
Video 0:07	The pouring hand has six fingers for a fraction of a second	2	Blocker	Trim the shot before the flaw or regenerate it
Video 0:09 to 0:12	A white cup in shot three becomes a blue cup in shot four	3	Fix	Specify the white cup in every shot description and regenerate shot four
Post image	The generator wrote "Sabah al-Mina" in disconnected, reversed letters	4	Blocker	Generate without text; add the headline in the editor using the brand font
Video 0:15	The voiceover pronounces the blend name correctly, but music masks the word "al-Mina"	6	Fix	Duck the music under speech
Video 0:17 to 0:20	"Order now" is on screen for only one second	7	Fix	Extend it to three seconds
Video 0:10	A background cup carries something resembling a well-known coffee chain's logo	5	Blocker	Blur the cup or regenerate with "no logos or brand marks"
Export	Captions sit at the very bottom, under the app's button area	9	Fix	Move them up to the lower-middle third

Decision: revise and re-review. Note for next time: for any physical product, start from a real photo and animate it; add "no text, no logos" to every prompt; Arabic text is always added in the editor.

After revision, the video was re-checked only on the sections that had findings, then published.

Common mistakes

Mistake: reviewing on a laptop only. Fix: always review on a phone at real size; that is where the audience sees it and where text is readable or not.
Mistake: the maker reviews their own piece straight after producing it. Fix: a second reviewer, or at least a short break and then a review by checklist rather than by impression.
Mistake: trying to fix Arabic text by regenerating again and again. Fix: generate without text and add it in the editor; faster and more accurate.
Mistake: accepting a product image that is "nicer than reality". Fix: comparison with the real product is a blocker, not a cosmetic item; the customer will compare on delivery.
Mistake: fixing one flaw and forgetting the fix can break something else. Fix: after each change, re-check the changed section and at least the two related to it (changing the music means re-checking timing, for example).
Mistake: not saving the prompt and source file. Fix: keep both with the final file; you will need them at the first revision request.

The 60-second pass before you hit "Publish"

When time is short, these items are never skipped:

The product shown is the product the customer will receive.
No extra fingers, no distorted faces, no real face without consent.
All Arabic text is readable and correct; every number, price and phone number matches the source.
No other company's logo; ours is the approved version.
Voice is clear above the music, and names are pronounced correctly.
Aspect ratio and platform are right; text is clear of the app interface areas.
The decision is recorded under the reviewer's name.