Wan 3.0 AI UGC Ad Generator
A UGC ad is four things at once — a creator, a room, your actual product and the voice that sells it — and here all four come out of one generation. No roster to pick a face from, no text-to-speech bolted on afterwards, and no lip-sync pass to marry the two.
Your first clip is free. No card.
Four cases from Wan's own published creator handbook, each printed with the prompt that produced it · Try this loads the prompt
TWO FACES, TWO VOICES · 16:9 · 15STwo faces, two voices, one pass
What a UGC ad generator does here
— and the one thing it does differently
A UGC ad is a person in an ordinary room saying something believable about a thing they are holding. Four elements, and the category builds them in three separate purchases.
Vertical, by default
Face, room, product and voice in the same pass
The usual stack goes: pick a face from a roster, send the script to a text-to-speech voice, then run a lip-sync pass to marry the two — and composite the product in afterwards. That is why so many of these share a particular deadness. The mouth is solving for a waveform that was recorded somewhere else, and the hand is holding something that was never in the room. Wan 3.0 samples all four together. The face, the kitchen, the bottle and the line arrive in the same render, which is also why turning the sound off does not make it cheaper — it was never a second product. The trade is real and it runs the other way too. There is no roster, so there is no face you can browse; you either describe who is on camera or hand over a photo. There is no variant button, so six hooks is six runs. And nothing carries forward between runs — two generations give you two people unless a reference image says otherwise. What a roster costs you sits on the other side of the same coin: a face you can pick is a face your competitor can pick. The four films above are Wan's own published handbook cases under CC BY 4.0, printed with their prompts verbatim so every number on this page can be checked against a file you can play.
- Longest single clip, natively multi-shot
- 30s
- Vertical at the landscape rate
- 9:16
- Reference images in one run
- 10
- Face, room, product and voice
- One pass
- Watermarks, on every tier
- 0
Longest single clip, natively multi-shot
Vertical at the landscape rate
Reference images in one run
Face, room, product and voice
Watermarks, on every tier
- No ad-account connection and no ROAS scoring — download the file and run it in your own manager.
- No avatar roster — the face comes from your reference photo, or from the sentence you write.
- No variant button — six hooks is six runs, at 40 credits each.
- No identity that survives between runs — put everything that has to match inside one thirty-second generation.
Four ways to cast a UGC ad, depending on what has to stay the same
All four run the same model at the same per-second rate. What changes is how much of the result you can ask for a second time — which is the only question that matters once one hook starts working and you need nine more like it.
Your SKU, in frame · up to 10 imagesHand over the product
One clean pack shot as a reference image, named in the prompt, and the thing in frame is the thing you sell rather than something shaped roughly like it. Set the ratio to 9:16 yourself — on adaptive, the framing is read from the photo you just uploaded.
- Duos and reaction ads · 5 audio refs
Cast two, and write the exchange
Two images, two voice references, and dialogue written as dialogue. The gym clip above is exactly this — two speakers, four lines, one generation — and the voice colour of each one is taken from an audio reference rather than picked off a list.
All four bill identically: 8 credits a second at 480P, 16 at 720P, 32 at 1080P. References are free — a UGC ad with ten reference images costs the same as one with none. Every limit and rate, per second.
A fifteen-second UGC ad holds twenty-nine words.
Your script is probably sixty.
This is the cheapest mistake to fix and the most expensive one to discover after the render. Both rates below are measured off Wan's own published cases, and they differ for a reason worth knowing: conversation has breathing in it, and a single read does not.
One creator, straight to camera
- Words a second
- 2.5
- 15 seconds
- 37 words
- 20 seconds
- 50 words
- 30 seconds
- 75 words
- Where the number comes from
- The handbook's dashboard case names its own rate in the prompt —
speaking speed about 2.5 words/second— and gets it
The default shape for a hook. Write to thirty-seven and the read still sounds unhurried.
- Reaction ads
Two creators, back and forth
- Words a second
- 1.93
- 15 seconds
- 29 words
- 20 seconds
- 38 words
- 30 seconds
- 58 words
- Where the number comes from
- The gym clip above runs 15.00 seconds and contains twenty-nine words across four lines — a turn of the head, a towel and a sip of water all cost time
Slower than it looks. Two people need room to react, and the reaction is what makes it read as UGC.
Voiceover over product footage
- Words a second
- 2.5
- 15 seconds
- 37 words
- 20 seconds
- 50 words
- 30 seconds
- 75 words
- Where the number comes from
- Same rate as a piece to camera — nothing is waiting for a mouth, but the pictures still have to land
Use when the product is the hero and there is no creator to cast.
Count the words before you count the seconds. A sixty-word script in a fifteen-second slot does not get politely trimmed — the read speeds up until it stops sounding like a person, which is the exact quality a UGC ad is buying. Cut to twenty-nine and the hook has room to breathe. Prompts with the clip each one produced.
Six ways a UGC ad stops being usable
None of these are about taste. Each one is a reason the file cannot be uploaded to an ad manager, which makes it a reason the credits are gone. Four of the six are visible in the clips above, against the prompts printed beside them.
Your vertical ad came back landscape
Ratio is a parameter, and its default reads your reference image. Upload a wide pack shot on adaptive and you have asked for a wide ad. The try-on clip above is 1280×720 for exactly this reason; the kitchen clip, with nothing uploaded, honours the 9:16 in its prompt.
Set 9:16 on the control before you press Generate. This page ships locked to vertical, so the only way to lose it is to change it.
Locked by defaultThe last beat never happened
Timecodes past thirty seconds are not clamped — they are discarded. The try-on prompt lays out four shots across ninety seconds and returns eight, with the beats compressed rather than the running time cut.
Keep the whole shot list inside thirty seconds and give the call to action its own window. If the spot is sixty, board it as two generations and cut them together.
Hard ceilingThe read is rushed, or a line is missing
More words than the seconds hold — twenty-nine in fifteen for a conversation, thirty-seven for one person reading to camera.
Count the script first, then pick the duration. Longer is cheap here: five more seconds at 480P is 40 credits.
ArithmeticHook two is a different person from hook one
Every generation is a fresh sample, and nothing about a face carries forward on its own.
Attach the same frontal still to every run and name it in the prompt, or put all the beats that must match inside one generation. How far the likeness actually holds.
Reference imageThe label on the pack is mush
Small type is where generative video is weakest, and a pack is mostly small type.
Shoot the legible frames as close-ups and say what has to be readable. Anything below headline size, plan a graphic over it in the edit. What the pack shot needs to look like.
Shot choiceThe take you liked has vanished
A finished clip is returned as a link that expires 24 hours after the run.
Download it, or run it signed in — a signed-in run is kept in your library instead of behind a link with a clock on it.
Save it
A UGC ad that errors out is refunded automatically. One that completes and is simply wrong is not — which is the whole argument for finding out at 480P, where the same model, the same cut and the same voice cost a quarter of the finished version.
A face, a claim, and two separate toggles
A UGC ad is a person appearing to recommend a product, which makes it the most regulated thing you can generate. Three rules decide whether your UGC ad runs, and none of them are about the model.
- A generated person is allowedThe FTC's Consumer Reviews and Testimonials Rule, in force since 21 October 2024, is explicit: "The rule has no blanket prohibition on the use of AI-generated avatars in marketing… Section 465.2 of the rule is drafted specifically so as to not prohibit companies from using virtual influencers."
- A generated customer is notThe line it draws is about what that person says: "It's possible that a company's use of such an avatar might be considered a 'testimonial' under the rule. That would be prohibited under the rule only if the underlying testimonials were fake or false." In a script, the difference is one verb — she can demonstrate the product and read a claim you can already substantiate; she cannot say she bought it and loves it, because nobody did.
- TikTok has a switchAds Manager carries a setting reading "This ad contains AI-generated content", required when an ad includes image, video or audio that is completely AI-generated or significantly modified by AI. Check it again on every duplicated campaign — duplication is how a hook test gets built.
- That is not the branded-content switchCommercial content disclosure is separate and answers a different question — this post promotes a product — and a Branded Content post needs it regardless of how the footage was made. Two settings, two problems, and missing either one costs distribution.
- Meta labels what it detects"Meta uses industry-standard detection methods, such as C2PA to identify when ad content has been created or edited using third-party generative AI tools." Read the current requirement on Meta's advertising standards before a launch — it is the platform's page to change.
- What you ownClips you generate are yours to publish, sell and bill a client for, on every paid plan, under Alibaba's usage policy for the Wan 3.0 model and our terms. No watermark and no separate licence to buy. Likeness, trademarks and any claim on screen stay your clearance problem, the same as with a camera and a real creator.
Upload someone's likeness or voice only with their agreement. A creator still of a real person is a reference the model will hold for thirty seconds, which is the feature and also exactly what makes misuse easy.
Six UGC hooks cost less than
one finished ad
Billing is per second of output, so length and resolution are the only two dials that move the number — the vertical cut is not a surcharge and the sound is not an add-on. A fifteen-second vertical UGC ad is 120 credits at 480P, 240 at 720P and 480 at 1080P. Six five-second hooks at 480P is 240 credits, which is half of one finished fifteen-second ad at 1080P. Test first, finish once.
What a UGC test costs, by shape
- Billed by
- output second
- One 5s hook, 480P
- 40 credits
- Six 5s hooks, 480P
- 240 credits
- One 15s ad, 720P
- 240 credits
- One 15s ad, 1080P
- 480 credits
- One 30s ad, 720P
- 480 credits
- Free on signup
- 1 clip + 24 credits
$39.90 a month on Pro is 2,240 credits — nine finished fifteen-second vertical ads at 720P, or fifty-six hook tests at 480P. Every figure above is the per-second rate times the seconds, so you can check the whole table with a calculator before you spend anything. Every plan and pack · what a failed generation costs.
兩種繳法，每月點數一樣
How to run a six-hook UGC ad test tonight
Five steps to a tested UGC ad. The first two cost nothing, and they are the two that decide whether the other three were worth running.
Write six openings, not one ad
Same offer, same product, six different first sentences. The hook that wins is reliably not the one you would have picked, which is the entire reason to generate rather than book.
FreeCount the words
Twenty-nine for a conversation, thirty-seven for one person talking, per fifteen seconds. A line that does not fit gets rushed rather than trimmed.
FreeRun all six at 480P, three seconds, vertical
240 credits for the set. You are looking for which opening survives being watched with the sound on, not whether the grade is right.
First passWatch them on a phone, muted, then unmuted
Muted finds the one that reads without a caption. Unmuted finds the one where the read sounds like a person. Different hooks win each test, and that is the useful part.
The honest passRe-run only the survivor, full length at 720P
One 15-second ad at 720P is 240 credits — the same as the whole test that found it. Attach the same reference image so the creator is the one you already validated.
Before you spend media budget
The order is the point. Every step above moves a decision earlier, to where changing your mind is still free.
AI UGC ad generator questions
What it takes in, what one ad costs, and what the platforms require before it runs.
What is a Wan 3.0 AI UGC ad generator?
Can I use my own product photo instead of a stock one?
Does it come with a voice, or do I add that afterwards?
How long can one UGC ad be?
Why did my vertical ad come out horizontal?
How many words fit in a fifteen-second ad?
Can I get the same creator across ten different UGC ads?
What does one UGC ad cost?
Is there a free way to try it?
Is there a watermark?
Can I run these as TikTok or Meta ads?
Can a generated creator say she loves the product?
Test the hook you are least sure about. It is free.
Take the UGC hook you cannot decide on, put your own product photo in it, and watch three seconds with the sound up. If it does not survive that, it will not survive a feed.
由 wan-3.run 編輯團隊撰寫與維護發布於 最後更新 工具 版本 2026.09.4