AI video for beauty tutorials: the same model face through before and after

The beauty-brand video market in 2026 is split between $29/mo avatar-presenter plans (HeyGen, Synthesia) and $90,000/mo Meta ad shops producing thirty avatar ads a month for $300 in tool credits (the Digital Applied case study). Neither price point addresses the format that beauty tutorials actually need: a before-process-after sequence where the SAME face must hold through every scene. An avatar presenter talks to camera for a minute; it does not render a model face in a bathroom mirror, then show the same face thirty seconds later with the finished look.

Face consistency across before and after is what the ranking "AI video for beauty tutorial" posts almost never price. The research (arXiv 2505.11425, the Face Consistency Benchmark) and the vendor pages (Kling's Omni, Pictory, Playcut) frame the problem in terms of models and benchmarks, not in terms of credits per tutorial.

Here is the credit math for a locked-face beauty tutorial series on our shelf.

The format this page prices

A beauty tutorial on social reads as a three-shot sequence: before (face bare, product in frame), process (hand applying product, often not the face), after (face revealed with the finished look). 15 seconds total, 9:16, voiceover over the top. The whole piece hangs on the before-face and the after-face being the same person — if they are not, the viewer treats the finished look as a different model entirely and the tutorial loses its credibility.

Avatar-presenter tools solve a different problem (one talking head, one scene). The locked-face solution our shelf serves is a three-image reference set — frontal, three-quarter, mid-shot — read by Kling O3 Pro as @Element1 into both the before shot and the after shot. The engine renders two framings of the same person, four seconds apart in the final cut, five minutes apart on the render queue.

The 15-second tutorial, priced

A single tutorial, 15 seconds, one locked face, three shots:

#	shot	engine	length	voiced?	credits
1	before — bare face holds product, voiceover opens	Kling O3 Pro 720p voiced	4s	yes	120
2	process — hand applying product, no face in frame	Kling O3 Pro 720p silent	6s	no	144
3	after — same face revealed, finished look, close-out line	Kling O3 Pro 720p voiced	5s	yes	150
—	15-second tutorial	—	15s	—	414

414 credits per tutorial. On Creator ($45 / 6,000 credits, 0.75¢ a credit) that is $3.11 per 15-second beauty tutorial. On Lite ($10 / 1,000 credits, 1.00¢) it is $4.14. On Studio annual (0.55¢ effective) it is $2.28.

The ranking SERP's cheapest comparable is the Digital Applied beauty-brand case: $300 in avatar credits for 30 ads, which is $10 per ad. The avatar ads are a different deliverable — a single talking shot, no before-after sequence, no face-reveal moment. Our shelf's $3.11 per tutorial is for the full three-shot sequence with the SAME face holding through all three.

The eight-tutorial month, on each plan

plan	monthly $	credits/mo	414-credit tutorials	per tutorial	surplus
Lite	$10	1,000	2	$5.00	172 cr
Starter	$18	2,250	5	$3.60	180 cr
Creator	$45	6,000	14	$3.21	204 cr
Studio	$99	15,000	36	$2.75	96 cr
Agency	$299	37,500	90	$3.32	240 cr

For an eight-tutorial month (one tutorial every three or four days), Creator at $45 is the plan: 3,312 credits spent, 2,688 credits of monthly headroom for a reference-set rebuild if the brand changes model, a longer 30-second feature piece, or a batch of behind-the-scenes clips. Starter at $18 handles five tutorials a month — enough for a brand posting Monday, Wednesday, Friday with a cadence break.

The face lock, built once

The lock is the whole page. Pick the model — the brand's spokesperson, a cast look that matches the demographic — and render a three-image sheet with the model's bare face (the before state) at three framings: full-frontal straight-on, three-quarter rotation, and mid-shot waist-up. Save all three to the project's reference library.

The one-time build on Kling O3 Pro at silent 24 cr/s:

step	engine	credits	$ on Creator
bare face, frontal, neutral lighting	Kling O3 Pro 720p silent	4s / 96	$0.72
bare face, 3/4 rotation	Kling O3 Pro 720p silent	4s / 96	$0.72
mid-shot, waist-up, neutral background	Kling O3 Pro 720p silent	4s / 96	$0.72
three-angle reference set	—	288	$2.16

288 credits once. Every tutorial for the quarter — or the year — reads from the same three images and renders the same face. The three images cover Kling's five-element budget (uses 3, leaves 2 open for a product shot or a brand prop) and Veo's three-ingredient cap exactly. For Seedance 2.5 you can expand the sheet to 30 images without rebuilding; most beauty briefs stay inside the three-image floor because the face is the only locked element.

Why the before and after need the SAME @Element1

Both the before shot and the after shot must prompt against the same reference index. On Kling O3 Pro that reads as:

Shot 1 (before, voiced):
@Element1 holds the tube in her left hand, bare face, bathroom-mirror lighting,
voiceover opens the tutorial, 4s, 720p.

Shot 2 (process, silent):
Close-up on hand applying product, no face in frame, same bathroom light, 6s, 720p.

Shot 3 (after, voiced):
@Element1 looks into the mirror with the finished look, same bathroom lighting
as shot 1, voiceover closes the tutorial, 5s, 720p.

The @Element1 token on shot 3 is pointing at the same reference image set as @Element1 on shot 1. The engine renders both from the same locked identity. Shot 2 does not reference the character because the face is not in frame — the model's hand is a non-face shot the engine can take without re-reading the reference, which is why we price it silent (and 20% cheaper than voiced).

Face Consistency Benchmark results (arXiv 2505.11425) report that 3–5 clean reference photos at frontal / 3-4 / profile angles hold identity through 12 seconds at 24fps in roughly 7 of 10 runs on reference-reading engines. Our shelf's three-image Kling pattern lands inside that range for the 15-second tutorial length.

The avatar meter is wrong for before-and-after

HeyGen's Avatar IV at $29/mo renders a single talking head in a single scene; the avatar does not change framing within the piece. The product-tutorial case Digital Applied documented (30 ads a month for $300 in credits) is one scene per ad, one framing per ad — not a before-process-after sequence.

The locked-face three-shot tutorial format needs an engine that:

1. Reads a reference set per shot (not a prebuilt avatar library).
2. Renders two different framings of the same identity (close-up before, close-up after).
3. Bills silent seconds cheaper (because the process shot does not include the face).

Kling O3 Pro clears all three; Synthesia's avatar library clears none. The per-minute avatar meter is a cheaper way to render 60 seconds of talking head than our 1,800 credits of voiced Kling. It is a MORE expensive way to render the three-shot tutorial, because the format needs multiple framings of the same person and the avatar library cannot deliver that from one preset.

The decision rule for beauty tutorials

Four 15-second tutorials a month, locked model face: Starter at $18/mo (one tutorial of headroom).

Eight 15-second tutorials a month: Creator at $45/mo (6 tutorials of headroom; room for a 30-second feature piece).

One daily 10-second product-shot avatar talking head (no face reveal): HeyGen Creator at $29/mo (per-minute meter beats us on single-framing).

30+ tutorials a month across a product org: Studio annual (0.55¢ effective credit; 36 tutorials on-plan).

Our free account carries 300 credits — enough to render one 10-second voiced Kling O3 Pro test (300 cr) to see how your lead model's face holds through a before-and-after. Build the three-angle face lock and render one tutorial on your product before you commit to a plan.

The $29/mo avatar presenter and the $300 ad-shop budget are the two prices the SERP uses to answer the question. The third price — $3.11 per 15-second locked-face before-and-after tutorial on Creator — is the one no beauty listicle publishes, because it is only possible on a shelf that bills per second across five engines with a reference-locked face carried through every shot.