SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsugc-testimonial

ugc-testimonial

General & other

View repository

Updated 10-15 22:19

About this skill

ugc-testimonial — UGC talking-head testimonial

The best-converting type of ad creative is often not a polished ad, buta recommendation that looks like a real person casually filmed it。

This skill generates that feel: handheld, home lighting, speaking like a real person.

1. Capability boundaries

Can do Description
Talking-head script Product + persona → a passage you can read aloud as-is
Storyboard Alternating structure of talking-head shots + product close-ups
UGC feel Slight handheld shake, indoor natural light, non-professional framing
Multiple personas Same product with different personas to produce multiple versions for A/B
Cannot do Description
Lip sync Generation models do not do lip sync well; for the finished video, use voiceover rather than sync sound
Impersonating a real person Do not claim it is a real buyer review, see "Compliance" below

2. Compliance first

UGC style ≠ fabricated reviews. The generated contentmust not claim to be a real buyer's real usage experience。

Allowed Not allowed
AI-generated demonstrative talking-head Impersonating a named real buyer's review
Scenario acting like "I wore this for a week" Fabricated positive-review screenshots, fabricated buyer shows
Label AI-generated as required by the platform Concealing AI-generated identity (some platforms mandate labeling)

Platforms have different labeling requirements for AI-generated content; confirm the target platform's rules before running ads.

3. Persona and script

Three elements of persona: who, what scenario, why credible.

Persona Opening line pattern
Commuting white-collar worker "Squeezing onto the subway every day, I need something that doesn't require upkeep..."
Student "With a budget of two hundred, I compared five stores..."
Mom "Taking care of the kid leaves no time to tidy myself up, so..."
Fitness crowd "After a workout I'm all sweaty, this one..."

Script skeleton (15 seconds)

0-3s   钩子：说一个具体的痛点，不说商品
3-8s   转折：为什么这个能解决 —— 给一个可验证的细节
8-12s  展示：镜头切到商品特写
12-15s 落点：给一个行动理由，语气随意

Three iron rules for writing talking-head scripts

  1. Say something specific. "Especially warm" is ineffective; "at minus five degrees I only wore this one" is effective.
  2. Leave hesitation. Real people pause and self-correct when speaking; all perfect long sentences feel fake.
  3. Don't recite specs. Put specs in subtitles; say feelings out loud.

4. Tool invocation

export DLAZY_VIDEO_MODEL=<你账号里可用的视频模型 ID>

node scripts/video.mjs --mode storyboard --task ugc-testimonial \
  --board examples/ugc-board.json --outdir out/ugc --subtitles

Alternate talking-head shots and product shots in the storyboard:

{
  "shots": [
    { "id": "talk1", "seconds": 4, "image": "assets/model/face-a.jpg",
      "prompt": "Handheld selfie-style shot, young woman talking to camera in a bright apartment, slight natural camera shake, warm window light, casual home background. She is wearing the olive cable-knit sweater.",
      "caption": "零下五度，我就穿了这一件" },
    { "id": "detail1", "seconds": 3, "image": "docs/flat-lay/garment-flatlay.jpg",
      "prompt": "Handheld close-up of the sweater texture, phone-camera look, natural indoor light.",
      "caption": "粗棒针织，比看起来厚" },
    { "id": "talk2", "seconds": 4, "image": "assets/model/face-a.jpg",
      "prompt": "Handheld selfie-style shot, same woman, she pulls the collar up and smiles, same apartment.",
      "caption": "洗了三次也没变形" }
  ]
}

Key prompt words：handheld、selfie-style、slight natural camera shake、 phone-camera look、natural indoor light、casual — these create the "unprofessional" feel. Do not write cinematic、professional lighting、studio, that would erase the UGC feel.

5. Execution flow

  1. Define the persona first. Ask the user who it is sold to; the persona follows the buyer profile.
  2. Write the script for the user to confirmSpoken script copy determines conversion more than the visuals.
  3. Face lockMultiple talking-head shots must feature the same person, use brand-kit for model.reference。
  4. --dry-run run through once, then do a real run.
  5. Add voiceover. Generated videos do not do lip sync; after the final cut, add a voiceover in editing software, it sounds more natural than sync sound.

Six, Common Issues

Symptom Cause What to do
Still looks like an ad The prompt contains cinematic / professional Change to handheld / phone-camera look
The shots are not of the same person No face lock Use --brand with model.reference
Lip movements do not match The model does not do lip sync Do not use sync sound, add a voiceover; or switch to handheld shots that do not show the face
Stiff expressions Faces are a weak point of video models Shorten the talking-head shots, use more product close-up cuts

Tips

  • Short talking-head shots, long product shots. The longer a face is generated, the more likely it breaks down; cut every 3–4 seconds.
  • Use the same script with different personas to make three versionsDo A/B testing, let the data from ad spend speak.
  • Subtitles are a must. Most people scroll through videos on mute.