Agent Clone · scale architecture

One still. A talking Albie. Anywhere.

The scalable default is generate-from-a-still: we make an AI photo of Albie in any setting or wardrobe, feed it his cloned voice + a script, and a model makes him talk — no filming, no fixed car clip. That's the path to thousands of videos a week. Below is the exact same still + same voice + same line, run through the two best still-to-talking-head models. Watch them in motion and pick the one that sells.

LoRA still · any setting→ Cloned voice · any script→ Avatar model · makes it talk→ QA + compliance→ Export

The source

never filmed

This photo doesn't exist. It's an AI still of Albie in a high-rise office he's never been in, wearing a suit we never shot — generated in seconds from his identity model. Every video below starts here.

100% generated · 0s to make · infinite variations

Albie · high-rise office · navy suit

Swap the prompt and he's on a beach, in a kitchen, on a podcast set, in a hoodie — same face, same person, new scene. That flexibility is the whole point: no scene needs a shoot.

The line

recruit · calm stage voice

"Every family I sit with is one bad day away from real trouble, and most have no plan. I get to be the person who shows up before that day, not after. That's the whole job."

Zero numbers, compliance-safe. Spoken in Albie Stage (his pick), dialed calm. Both videos use this identical audio track — so the only variable you're judging is the model's motion + lip-sync.

Head to head

same still · same voice · same line

A · Kling Avatar v2 Probeard specialist

1232×1648 · 30fps~$1.70/15srender ~4 min

Benchmark leader on bearded/detailed faces. In our frames it added a natural hand gesture — more of the "spit facts" energy Albie wanted.

B · OmniHuman 1.5film-grade

1248×1664 · 25fps~$2.40/15srender ~5 min

ByteDance's cinematic model. Subtle camera movement and a very natural resting face — reads more "shot on a real camera."

How to judge — watch in motion, not stills

Both look great frozen. The lesson from the lip-sync bake-off: a still can't tell you if it's synced. Play each full, sound on, and rate:

Lip-sync lock — do the lips actually hit the words, or does it drift into "voiceover over a moving face"?
Believability — could a stranger scrolling think this is a real man talking to a phone?
Energy — does it carry conviction, or go slack and dead-eyed?
Identity — still unmistakably Albie the whole way through?

	A · Kling v2 Pro	B · OmniHuman 1.5
Cost / 15s	~$1.70 (cheaper)	~$2.40
At 2,000 vids/wk	~$3.4k/wk	~$4.8k/wk
Motion feel	Gestures, expressive	Cinematic, restrained
Best when	UGC / talk-to-camera	Polished / brand hero

Tell me A or B and it becomes the studio's default talking-head engine. We can also keep both — Kling for volume UGC, OmniHuman for the occasional hero cut — and let the mode pick. Once you call it, I wire it into the live Create flow so every reel renders this way, at scale.