November 2025
Most AI video produced today is recognisable in the first three seconds. The reason is not the model. It is the absence of editing judgment, story decisions and brand-fit work. The model is a generator, not a director. Reel uses state-of-the-art generators every day, and has not shipped a single film that reads as AI-made. Here is the workflow that separates the two.
01. What the AI layer does
- Style-reference frames at script stage — generated thumbnails per beat, so the director sees the film before anything is shot
- Motion drafts — image-to-video turns key frames into 6-second clips that previs the cut; a week of junior-editor time becomes a morning
- B-roll and atmospheric inserts — weather plates, establishing shots, mood inserts; anything without a face in close-up
- Scratch tracks — AI voice for the offline edit; the real voiceover comes from a human actor later
- Asset variations at scale — one hero composition, twelve aspect ratios, six language overlays; only the winners ship
02. What the human layer does
- Story structure and beat selection — which moments earn the cut
- Casting and direction — the camera always rolls on real actors for anything in close-up
- The cut — pacing, breath, where to hold a beat, when to land the cold cut
- Colour grade — human-graded on a real reference monitor
- Sound design — foley, room tone, score selection, mix
- Brand-fit pass — does this film read as that brand, in that market, to that audience
Anything that touches the final ten percent, where craft lives, stays human. The AI layer never approves anything.
03. Per-client tuning
There is no fixed ratio. A luxury brand wants 100% live action with AI assist on style references only. A B2B SaaS brand might want 30% live — the CEO to camera — and 70% generated b-roll. A category-creator brand may want stylised, AI-first work. A regulated-industry client wants the provenance of every frame disclosed, in a workflow that survives audit. We mix the workflow to fit the brand, not the other way around.
04. Disclosure
Every project’s master brief contains a workflow disclosure: what percentage of frames came from generation versus camera, where AI was in the chain, who graded final, who voiced the final track. It is not legally required in most jurisdictions. It is just honest.
“Brands that want to claim “100% no AI” do not engage Reel. Brands that want to ship low-effort fully-AI content also do not engage Reel. The middle wants the truth on file.”
05. Where it still breaks
- Faces in close-up — anything within thirty frames of a face has to be a real human in front of a real camera
- Product hero shots — real product, real lens, real lighting; generated hero shots get rejected in client QA every time
- Narrative coherence past 60 seconds — longer pieces are cut from shorter generations, with a human deciding how they stitch
These are current ceilings, not permanent ones. We rebuild the workflow every quarter against what is genuinely possible now.
06. The principle
AI is raw stock. Editing is craft. Mixing the two is the actual work. The deliverable is a film with the speed of AI and the taste of a human cut — neither alone gets there.
End · Entry №02 · Filed November 2025
