The AI Video Spec Cheat Sheet: Reels, Shorts, Product Pages, Marketplaces
By the ORA Lab team · Updated 9 September 2026 · 8 min read
Key takeaways
- Product video is one asset wearing five uniforms: generate a single high-quality master per SKU, then cut, crop, and grade per placement, never generate per placement.
- The master recipe: landscape or square, 8 to 12 seconds, natively sharp, true-colour graded, with a loop-friendly ending, every downstream format cuts cleanly from it.
- Aspect ratio is the big sort: 9:16 for Reels/Shorts/Stories, 1:1 or 4:5 for feeds, 16:9 for product pages and YouTube, and vertical crops need the product framed in the centre third.
- Platforms recompress everything: assume your clip will be re-encoded, so protect it with sharp sources, restrained motion, high-contrast text, and colour that survives compression.
- Specs drift, platforms change caps and dimensions quietly, so verify current numbers in each platform's help pages before a big launch; the workflow here outlives any single spec.
Somewhere on every marketing team's drive is a folder of good clips that never shipped, wrong ratio for Reels, two seconds over a marketplace cap, text baked into a corner that Stories chrome covers. Spec friction is boring, which is why it quietly kills more video distribution than creative quality does. And it's doubly wasteful in a generation workflow, where teams sometimes regenerate a clip per placement, paying video generation prices for what is, in every professional pipeline, an editing task.
This is the reference sheet we wish every brand team had pinned: what each placement wants, why the one-master workflow feeds them all, and the handful of production rules that make a clip survive every platform's recompression. One honest caveat up front, platforms adjust caps and dimensions quietly and often, so treat the numbers here as the stable 2026 shape of things and verify current specifics in each platform's help pages before a major launch. The workflow is the durable part.
The one-master workflow
Everything downstream gets easier if the master is right. The recipe: landscape (16:9) or square, because width can be cropped to vertical but vertical can't grow sideways, with the product or scene's key action framed in the centre third, so the 9:16 crop finds it; 8 to 12 seconds, which trims down to every cap that matters; natively sharp at your largest target resolution, per the resolution rules; true-colour graded so gold stays gold after re-encoding; and an ending that hands back to the opening frame, because half the placements below loop. Generate two or three takes of this master per the standard workflow, pick the keeper, and the rest of this article is an editing checklist, not a generation task.
The placement matrix
| Placement | Ratio | Length that works | Sound | Loop? | Notes |
|---|---|---|---|---|---|
| Instagram Reels / YT Shorts | 9:16 | 7 to 15s for product clips | Off-first design | Auto-replays | Centre-third framing; safe zones top & bottom for UI chrome |
| Stories | 9:16 | ≤15s per card | Off-first | No | Tappable pace; one idea per card |
| Feed (IG/FB) | 4:5 or 1:1 | 8 to 15s | Off-first | Auto-replays | 4:5 buys more screen than 1:1; hook in first 1.5s |
| Product page (PDP) | 16:9 or 1:1 | 6 to 10s loop | Silent | Seamless loop | Autoplay muted; the breathing-scene loop is the conversion placement |
| Paid ads (Meta) | 9:16 + 4:5 variants | 8 to 15s | Off-first | Auto-replays | Upload placement-specific ratios; the ad structures playbook applies |
| Marketplaces (Amazon etc.) | 16:9 typical | 15 to 60s caps vary | Optional | No | Product-only content rules; check each platform's video policy |
| YouTube (long placements) | 16:9 | As needed | On | No | Assembled edits of clips, not raw generations |
Surviving recompression
- Assume re-encoding: every platform recompresses uploads, some brutally. Upload the highest-bitrate version the platform accepts and let it do the damage from a good source rather than a mediocre one.
- Restrained motion compresses better: the slow drifts and breathing light that flatter products anyway also produce fewer compression artefacts than fast pans, elegance and bitrate agree here.
- Fine detail needs size headroom: pavé fields and fabric weaves smear first under compression; if the placement is small (feed thumbnails), favour the tighter crop over the wide scene.
- Text must be chunky and contrasty: thin elegant type dissolves in re-encoding; short high-contrast overlays inside safe zones survive. One idea per overlay, two seconds minimum on screen.
- Preview on a phone from the actual platform: the only honest QA for compression is the placement itself, what your monitor shows is not what the feed serves.
Naming and the asset library
Spec compliance decays without a system: three months in, nobody remembers which file was the PDP loop and which was the ad hook. The fix costs one convention, SKU_placement_ratio_duration (MER3S_pdp_16x9_8s, MER3S_reel_9x16_12s), applied when the cuts are made, plus one rule: masters live separately from cuts, because next season's formats get cut from the master, never from a cut. This is the same systemise-once discipline that governs looks and motion language, applied to files, dull, five minutes, and the difference between a video library and a video junk drawer. It also makes the fortnightly ad-rotation rhythm frictionless, because variants slot into a naming pattern instead of a guessing game.
The pre-launch checklist
- 1.Master approved: product frame-level QA'd, true-colour graded, loop point clean.
- 2.Cuts made per the matrix: vertical with safe zones respected, feed crop, PDP loop trimmed to its smoothest stretch.
- 3.Current specs verified: thirty seconds in each target platform's help pages, caps and dimensions drift quietly.
- 4.Phone preview from each placement: compression, colour, and text legibility checked where the customer actually sees them.
- 5.Files named and filed: masters archived, cuts labelled by SKU_placement_ratio_duration.
None of this is glamorous, and all of it is why two brands with identical creative see different results in distribution. The generation side keeps getting easier, one photo becomes a clip in an afternoon, which makes the spec layer the remaining place velocity leaks. If you'd rather the whole chain arrive handled, master, cuts, safe zones, naming, that's how ORA ships video to clients: placement-ready sets, not raw clips. and ask to see a delivery folder; the boring parts are where the time goes back.
Frequently asked questions
- What are the video specs for Instagram Reels product videos?
- 9:16 vertical, with 7 to 15 seconds the working range for product clips, designed for sound-off viewing with the product framed in the centre third and text inside safe zones (clear of the top ~15% and bottom ~20% where UI chrome sits). Reels auto-replay, so loop-friendly endings earn free second views. Verify current caps in Instagram's help pages before launch.
- Should I generate separate videos for each platform?
- No, generate one high-quality master per SKU (landscape or square, 8 to 12 seconds, centre-third framing, loop-friendly ending) and produce every placement by cropping, trimming, and grading in an editor. Regenerating per placement pays generation prices for editing tasks and multiplies your QA burden.
- What video length works on a product page?
- A 6 to 10 second seamless loop, autoplaying muted near the add-to-cart area. The breathing-scene loop, subtle ambient motion around a stable product, is the placement most associated with shoppers lingering and converting, and it needs a clean loop point more than it needs length.
- Why do my product videos look worse after uploading?
- Platform recompression: every network re-encodes uploads, punishing fast motion, fine detail, and thin text first. Protect clips with sharp sources, slow restrained motion, chunky high-contrast overlays, and the highest-bitrate upload accepted, then QA on a phone from the actual placement, which is the only honest preview.
- Do marketplaces like Amazon allow AI-generated product videos?
- Marketplace video programs generally accept any footage that meets their content rules, product-focused, accurate, within length and format caps (commonly 16:9, with varying duration limits). The accuracy bar is the real constraint: the video must truthfully depict the sold product, so frame-level fidelity QA matters more than the generation method. Check each platform's current video policy.