AI as raw stock, not finished cloth

Most AI video produced today is recognizable 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 single day, and we have not shipped a single film that reads as AI-made. Here is the workflow that separates the two.

§01What the AI layer does

The generation stack is doing real work, just not the work that traditional shops imagine when they hear “AI video.”

• ·Style-reference frames at script stage. Generated thumbnails per beat so the director can see the film before anyone reshoots a thing.
• ·Motion drafts. Image-to-video models turn key frames into 6-second clips that previs the cut. This used to take a week and a junior editor; it now takes a morning.
• ·B-roll and atmospheric inserts. Weather plates, location establishing shots, mood inserts. Anything that does not have a face in close-up.
• ·Scratch tracks. AI voice for the offline edit. The real voiceover comes from a human actor later; the scratch lets us cut to timing before we hire anyone.
• ·Asset variations at scale. One hero composition, twelve aspect ratios, six language overlays, all generated. Only the winners ship.

§02What the human layer does

Everything that touches taste.

• ·Story structure and beat selection. Which moments earn the cut.
• ·Casting and direction for human-performed pieces. 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.
• ·Color grade. Final color is 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.

§03Per-client tuning

There is no fixed ratio. Every brand needs a different mix.

• ·A luxury brand wants 100% live action with AI assist on style references only.
• ·A B2B SaaS brand wants 30% live (their CEO talking to camera) and 70% AI-generated b-roll, because the AI material looks more like their product than reality does.
• ·A category-creator brand wants stylized AI-first because their brand is itself surreal.
• ·A regulated-industry clientwants disclosure of every frame's provenance and a workflow that survives audit.

We mix the workflow to fit the brand, not the other way around.

§04Disclosure

Every project's master brief contains a workflow disclosure. The client knows exactly what percentage of frames came from generation versus camera, where AI was in the chain, who graded final, who voiced the final track. This 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.”

§05Where it still breaks

We are honest about what the generation stack still cannot do well.

• ·Faces in close-up. Anything inside thirty frames of a face has to be a real human in front of a real camera. The uncanny valley is still real, and it punishes the spend.
• ·Product hero shots. Real product, real lens, real lighting. Generated hero shots get rejected by clients in QA every time.
• ·Narrative coherence past 60 seconds. Generators drop continuity past about a minute. Anything longer is cut from multiple shorter generations, with a human deciding how the pieces stitch.

These are not permanent ceilings. They are current ones. We rebuild the workflow every quarter against what is genuinely possible now, not what was possible a year ago.

§06The principle

AI is raw stock. Editing is craft. Mixing the two is the actual work. The deliverable is a film that has the speed of AI and the taste of a human cut. Neither alone gets there. The combination, run honestly, is faster and better than either.