Skip to content

Use cases · AI UGC ads

UGC-style ads you can test at scale

A talking presenter from one photo, a voice from ElevenLabs, a script and captions: one ad, then as many hook variants as you want to test. Every ad has a price you can read before you render it.

From $2.25 per piece at current prices

Looking for the studio flow, from a URL to a finished ad? See AI UGC ads

What you can make

  • Talking-head ads from a single photo and a voice
  • Hook variants of the same ad for creative testing
  • Product demos with b-roll cut between the presenter's lines
  • The same ad dubbed into other languages
  • 9:16, square and 16:9 versions with reframe

The pipeline

  1. 1

    Script and hooks

    One script run writes the ad; ask for several openings to test against each other.

  2. 2

    Voice

    ElevenLabs reads each version. A 20 second read fits in one block of 450 characters.

  3. 3

    Presenter

    The avatar model turns one photo and the voice into a talking-head video, billed per second.

  4. 4

    Cutaways and captions

    Add Seedance or Veo 3.1 Fast b-roll between lines, burn in captions and export for each placement.

Cost per piece

Each line is a real row of the rate table at the public prepaid price, multiplied out. A range means the price depends on something only the run decides, such as the length of the text. Each job shows what it actually cost.

A 20 second talking-head ad

Avatar video billed per second, one voice block.

20 × seconds of avatar video$0.11 per second$2.14
1 × voiceover blocksElevenLabs · $0.068 per 450 characters$0.068
1 × script run$0.038–$0.095 per run$0.038–$0.095
1 × captions run$0.009–$0.038 per run$0.009–$0.038
1 × assembly and exportour own compute · $0 per run$0
Total · at current prices$2.25–$2.34

Five hook variants (× 5): $11.26–$11.69 at current prices.

Every rate, machine-readable: GET /api/pricing

Run it from Claude Code

Connect the hosted MCP server once, then describe the batch in plain words. Claude plans it, calls the models, waits for each render and hands you the files with what they cost.

claude mcp add --transport http aitachyon \
  https://aitachyon.com/api/mcp \
  --header "Authorization: Bearer ait_..."

“Write five hooks for our magnesium sleep gummies, then make a 20 second UGC ad for each with the presenter photo I uploaded, a warm ElevenLabs voice and captions, 9:16. Show me the cost per ad first.”

Or from your code

Plain HTTP with a bearer key. Video returns a job you poll; every file carries a ref you can fetch later with GET /api/generations/{ref}.

curl https://aitachyon.com/api/voiceover/generate \
  -H "Authorization: Bearer ait_your_key" \
  -H "Content-Type: application/json" \
  -d '{"projectId":"<project id>","provider":"elevenlabs","voiceId":"<from GET /api/voiceover/voices>","text":"I stopped buying three different serums the week I found this one."}'
curl https://aitachyon.com/api/avatar/generate \
  -H "Authorization: Bearer ait_your_key" \
  -H "Content-Type: application/json" \
  -d '{"projectId":"<project id>","voiceoverId":"<voiceover.id>","audioUrl":"<voiceover.audioUrl>","imageUrl":"https://example.com/presenter.jpg","model":"kling"}'

Built for volume

  • Pay per call from a prepaid dollar balance that never expires
  • A failed render is refunded to your balance automatically
  • Per-key spend, alerts and one-click revoke, so a runaway script stays contained
  • Every file has a ref you can fetch with GET /api/generations/{ref}
  • The web studio, the HTTP API and the MCP server share one account and one balance

Platform rules you need to respect

Generating is the easy part. What you publish is your responsibility, and each platform has rules for AI content. Never present an AI person as a real customer, a real review or a real event.

YouTube

Monetized channels are reviewed for inauthentic content: mass-produced or repetitive videos without real value added. Realistic altered or synthetic content has to be disclosed when you upload.

TikTok

Realistic AI-generated content has to carry a label, set with the AI-generated content toggle when you post. TikTok may also label it automatically.

Meta (Facebook, Instagram)

Meta labels content it detects as AI-made and asks you to disclose photorealistic video and realistic audio created or altered with AI. Ads about social issues, elections or politics have their own disclosure rules.

Policies change: read the current version before you publish. We make the files; we do not promise monetization, reach or views.

Questions