Skip to content
Yito.ai
Internal benchmark · not a customer

Beauty Product Photo Case Study: Serum Reveal

For Beauty ecommerce teams evaluating a controlled product reveal: start with one owned editorial serum photo and a preservation-first motion brief.

Acceptable result: a four-second review candidate with the retained source, settings, output, observed failures, and human decision.

Input
720 × 900 serum photo
Model
Hailuo 2.3 Fast
Output
768 × 960, 4 seconds
Studio estimate
113 Credits

Workflow contract

Process

  1. Record the source and identity-critical package features.
  2. Generate one continuous reveal with no invented claims or text.
  3. Inspect multiple timestamps and retain the acceptance decision.

Quality check

  1. Accepted: bottle, dropper, stone set, warm light, and quantity.
  2. Not proven: small label text, claims, repeatability, rights, or channel approval.

Delivery

  1. product-page media candidate
  2. short social insert

Models and tools

  1. Image to Video
  2. Product Video Ad

Credits

The internal evidence run charged no user account; the equivalent Studio route estimated 113 Credits.

Direct answer

This beauty case turns one approved-looking amber-serum source photo into a restrained product reveal. Hailuo 2.3 Fast received a slow-orbit brief and returned a four-second clip. Review accepted the bottle, dropper, stone set, warm highlights, and single-product count, while small label text and the duration mismatch remain explicit limitations.

Verified production outputs

Generation cases retain the input, settings, and historical cost; local audits retain the inspected file and every observed check.

Source image of an amber serum bottle on a dark stone setInput
Output

Amber serum product reveal

Model: Hailuo 2.3 Fast via Replicate

The accepted output keeps one amber bottle, the dropper, dark stone setting, and warm highlight direction recognizable through the clip. No new logo, product claim, hand, or second bottle appears.

The small label area cannot prove character-perfect text preservation, and the delivered four-second artifact is shorter than the six-second request recorded by the editorial run.

Prompt motion
Slow orbit, restrained atmosphere
Requested duration
6 seconds
Delivered artifact
4-second MP4 + WebM
Audio
None
Historical Credits
113

Observed failures

  • Reject a retry if the dropper, bottle shoulder, label area, material, or product count changes.
  • Small printed packaging text still needs a frame-by-frame comparison against the approved pack artwork.
  • The atmospheric particles and reflections move synthetically and should not be presented as ingredient evidence.
Open the full guide, FAQs, and sourcesUse the detailed method when you need it without extending the tool by default.

Case brief and source condition

The source shows one amber dropper bottle on a dark stone surface, with warm highlights and enough separation to read the silhouette. The editorial goal was not to redesign the pack or add a beauty claim. It was to introduce controlled movement that could support a skincare product-page hero, launch section, or short social edit. The retained prompt asked for a slow camera orbit, realistic reflections, faint atmospheric movement, preservation of the bottle shape, dropper, material, label area, and composition, one continuous shot, and no cuts, text, or invented logos.

Model, settings, timing, and Credits

The run used Hailuo 2.3 Fast through Replicate. The editorial record requested six seconds and logged 117,020 milliseconds of provider-total generation time. The available production artifact is four seconds in MP4 and WebM, with a 768 × 960 retained final frame and a 113-Credit Studio estimate. The editorial run itself charged zero user Credits because it was created as owned launch evidence, so 113 is an estimate for the equivalent Studio route, not a claim that a customer account was billed. A new request must use the quote shown before submission.

What the human review accepted

Review compared the source and motion result rather than judging the poster in isolation. The output keeps one amber bottle, the recognizable shoulder, dropper, dark stone setting, and warm-light direction. It does not introduce a hand, second product, promotional caption, obvious logo, or new package claim. The orbit and atmosphere remain restrained enough for the product to stay central. This is useful evidence that one specific source and prompt produced a viable candidate. It is not a general score for all skincare packs, glass, labels, or Hailuo retries.

Observed failures and rejection rules

The largest retained mismatch is duration: six seconds were requested, but the artifact available for publication is four. The small label area cannot demonstrate character-perfect preservation, so a real branded serum would require frame-by-frame comparison with approved packaging artwork. Reject any retry that changes the dropper length, bottle shoulder, glass color, fill implication, label geometry, cap, material, product count, or set contact. Atmospheric particles and moving reflections are decorative; they cannot be cited as evidence of an ingredient, texture, absorption rate, clinical result, or other performance claim.

Where this asset fits and does not fit

The accepted clip can support a product-page media gallery, launch header, collection story, or a short transition inside a larger social edit, provided the theme and placement accept its format. It should not replace a factual primary pack shot, an ingredient diagram, a before-and-after claim, regulatory labeling, or a close-up intended to prove printed text. For TikTok Shop, assemble it inside the final vertical edit with safe zones, subtitles, approved offer copy, audio rights, and the exact linked SKU. For Shopify, test the real product-media component and select a deliberate poster.

How to reproduce the review discipline

Start with an owned or licensed source that exposes every identity-critical feature. Write the preservation list before the motion instruction, generate one candidate, and inspect multiple timestamps at full size. Record model, provider, request, delivered dimensions, duration, quote, output paths, reviewer, date, accepted features, failures, and planned channel. Keep the rejected reasons with the recipe so the next brief can improve. If small text, glass geometry, or product truth cannot be verified, combine conventional photography with generated atmosphere instead of asking the model to reconstruct the regulated area.

Explore next

Limits and review ownership

This is one owned editorial run. It does not prove label accuracy, cosmetic claims, repeatability, rights, channel approval, or future price, and its delivered duration differs from the request.

Move from evidence to production

Open the campaign plan or Studio, review the current quote, and use pricing only when the workflow fits.