Credits burn on failed generations
The loudest complaint in the category. Failed renders consume credits, unused credits expire at the end of the month, and refunds are refused even for technical failures.
Turn one product photo into a UGC ad video that ships with its own production record and a two-part AI disclosure. Every stage is written down. You are only charged for output you actually receive.
Production certificate · excerpt
Paid advertisement for Northlight. Presenter is AI-generated. Not a real customer or endorsement.
Sound on. The overview walks the pipeline and ends on the production certificate, not on a beauty shot.
What this category does to you
Four complaints repeat across every review platform and every agency teardown of AI UGC tooling. They are not edge cases, they are the category's operating model.
The loudest complaint in the category. Failed renders consume credits, unused credits expire at the end of the month, and refunds are refused even for technical failures.
When a render fails or looks wrong there is no prompt, no model name, no cost and no failure point. One widely shared agency teardown described a market leader as an engine with no strategy layer at all.
Pricing pages behind a login or returning 404. Charges continuing after cancellation. Annual plans entered by default. Trustpilot scores across the category sit between 2.3 and 2.8.
A synthetic person appearing to endorse a product is the category's core output, and it is the exact artefact regulators treat as deceptive when presented as a testimonial. Almost every tool ships it with no guardrail and no record.
The pipeline is thirteen stages. You do not have to care about that, until something goes wrong and suddenly you do.
512px minimum, checked by magic bytes rather than the filename. Add the brief: audience, objective, platform, tone, call to action.
Each stage reports what it did, what it skipped and why. Prompts, model names and durations are written down as they happen, not reconstructed afterwards.
A vertical video with the disclosure in frame, plus the production record: disclosure text, model provenance, lint findings and per-stage timings.
The glass box
Most tools report one green tick for the whole pipeline. This one reports per stage, and a stage that has not been built yet is marked skipped rather than complete.
That honesty is the point. If you cannot see which stages ran, the record is worthless the moment anyone asks a hard question about the ad.
run_5f2a · production record
delivered| stage | state | detail | time |
|---|---|---|---|
| product-input | ran | 1080x1080 · sha256 verified | 0.2s |
| product-analysis | skipped | not implemented yet | — |
| campaign-strategy | skipped | not implemented yet | — |
| ugc-script | ran | lint: 0 blocking, 1 warning | 0.4s |
| storyboard | ran | 4 scenes | 0.1s |
| presenter-generation | skipped | not implemented yet | — |
| scene-image-generation | ran | higgsfield · gpt_image_2 | 31s |
| product-placement | skipped | not implemented yet | — |
| video-generation | ran | higgsfield · seedance_2_0 | 2m 48s |
| audio-generation | skipped | not implemented yet | — |
| video-composition | ran | 9:16 · 720p confirmed | 0.1s |
| quality-review | ran | record written | 0.2s |
| final-output | ran | asset published | 0.1s |
Five stages are marked skipped because they are not implemented yet. We would rather show you that than pretend otherwise.
The part nobody else builds
Guidance in 2026 is direct about this. AI-involved sponsored content needs a double disclosure, the paid relationship and the AI involvement, each satisfied independently. A platform “made with AI” label does not cover the commercial relationship, and an ad hashtag does not cover the AI. Fabricated testimonials are treated as deceptive whether or not they carry a label.
So the studio refuses to write them. Scripts are demonstration-led, first-person endorsement language is blocked before generation, and the disclosure goes into the frame rather than into an upload description that a repost strips.
First-person use claims, personal result claims and personal endorsements are rejected at the script stage. "I have been using this for weeks" never reaches the model.
Whenever a synthetic presenter appears, the disclosure is a persistent lower-third caption rather than metadata. Muted autoplay and reposts keep it.
Disclosure text, disclosure mode, lint findings and model provenance are stored per ad with a timestamp, so you can produce the record months later.
To be explicit, because this matters: UGC Genie generates a disclosure and keeps a record. It does not certify your advertising as lawful, and nothing here is legal advice. You and your advisers remain responsible for the claims you make.
Pricing
No login wall to see a number. Credits are consumed by delivered output only.
$0to start
One free production, then bring your own Higgsfield key. Unlimited, no credits.
$49per month
For a brand shipping creative every week.
$149per month
For teams answering to someone else's legal department.
Ask these of anyone in this category, including us.