How to Make UGC Ads With AI in 30 Seconds

Key takeaways

A UGC video that cost about $250 and a two-week wait from a human creator now generates in about 30 seconds (Alec Saluga, Session 008).

Meta's algorithm demands an insane volume of creatives, which is exactly what per-variant AI generation supplies.

The Marketing Studio recipe is three uploads: a prompt, the product image, and an avatar.

Nobody at scale targets manually anymore. Advertisers spending tens of thousands a day run broad and let the creative do the targeting.

The same avatars work beyond paid ads: organic social, AI influencers, and VSL-style videos explaining your product.

What changed about ad creative?

Meta's algorithm rewards advertisers who feed it a constant stream of new creative, and Alec Saluga's words for the appetite were blunt: "The algorithm demands an insane volume of creatives." Before AI, each unit of that volume was a roughly $250 invoice to a UGC creator and a two-week wait, with revisions costing another two.

AI generation collapses both numbers. The cost story is the whole reason this workflow exists, and it is why the testing motion below became practical for businesses that could never afford creator volume.

	Human creator (before)	AI generation (Session 008, live)
Cost per video	About $250	Effectively free (subscription and tokens)
Turnaround	Two weeks, plus two more for revisions	About 30 seconds per ad
Revisions	A request and another wait	Regenerate with a tweaked prompt
Volume	A handful per month	As many hooks as you can test

How do you actually make one?

Alec ran the full cycle live on Session 008 with an AI4NTP dad hat that did not exist ten minutes earlier. The product image came from one line of prompting ("make a cool dad hat with this logo, give me the product image"). The image went into Higgsfield's Marketing Studio with an avatar and a script prompt drafted with Claude, and the finished ads played on the call.

The first generations were unpolished on purpose. Alec had not QA'd the avatars, and the scripts needed a stronger call to action, but that is the point of the workflow: a weak variant costs 30 seconds, not $250, so you fix it by generating again.

Are AI UGC ads believable yet?

Believability is close, and closing fast. The hat in the generated ads was the actual product image, worn on camera by an avatar, and Alec's live assessment was that the output is "essentially indistinguishable, or at least well on the way to getting there," with end of year as his deadline for the gap disappearing. The data side is less ambiguous: the ads are already working for his own campaigns and his clients'.

Do you still need audience targeting?

Manual targeting is effectively over at scale. Answering a live question, Alec said everyone he knows, including accounts spending tens of thousands of dollars a day, runs broad: "Your creative's doing the targeting at this point." The volume of hooks and scripts you can now generate is precisely what makes broad viable. Retargeting site visitors remains its own, different motion.

Where else do AI avatars work?

Paid ads are the start, not the boundary. Alec pointed at the rise of AI influencers for organic social, and at VSL-style videos, a YouTube-length explainer of your product, as the same workflow with a longer script. His rule of thumb: any marketing asset where a person on camera makes sense is now an asset AI can produce faster, cheaper, and in more variants.

The repeatable order of operations

If you take one thing, take the sequence. It works because each phase feeds the next.

Generate the product image

One line of prompting is enough to start: name the product, attach the logo, ask for the product image. Alec's exact prompt for the demo hat was "make a cool dad hat with this logo, give me the product image."

Start with Higgsfield

Draft the script with an LLM

Have the model write the hook and script for the avatar to speak. Alec used Claude for the demo script, then noted the obvious next iteration: a sharper call to action per variant.

Start with Claude (Anthropic)

Assemble the ad in Marketing Studio

Three uploads: the prompt, the product image, and an avatar. Higgsfield aggregates the major image and video models under one interface, so you can toggle models without changing workflow.

Start with Higgsfield

Generate, review, regenerate

First generations arrive in about 30 seconds each. Expect them rough, fix by regenerating rather than polishing: weak hook, new prompt; weak avatar read, new take.

Run broad and let the data pick

Ship many hooks, target broadly, and let per-creative performance decide. At current CPMs, creative volume is the targeting.

Frequently asked questions

How much do AI UGC ads cost compared to creators?

A creator UGC video ran about $250 with a two-week turnaround, and revisions added another two weeks. The AI version generates in about 30 seconds for roughly the cost of a Higgsfield subscription and model tokens, which is what makes testing dozens of hooks practical.

Are AI UGC ads good enough to use?

Alec Saluga's live read on Session 008: essentially indistinguishable or well on the way, there by end of year for certain. More practically, the ads are already performing in his and his clients' campaigns, and every variant that is not good enough costs 30 seconds to replace.

Do I still need to set up audience targeting?

Not the manual kind. Advertisers spending tens of thousands a day run broad and let the creative do the targeting, which AI-generated volume makes possible. Retargeting your own site visitors is still a separate, worthwhile motion.

Where this came from

Session 008 · Recorded live

Build and grow a company with AI in under an hour.

Watch the recap →YouTube replay →

21:00 AlecThe algorithm demands an insane volume of creatives.

22:00 Alecyou know, $250 for a, you know, creator to make a UGC video for you. And now, that's also achievable with AI in seconds, for free.

22:00 Alecmy prompt engineering here, real advanced prompting techniques, make a cool dad hat with this logo, give me the product image

23:00 Aleca prompt, I could upload the product image, I can upload an avatar

24:00 AlecThis took 30 seconds to generate, I threw a quick prompt in there that I used Claude to help generate.

24:00 AlecIt's essentially indistinguishable, or at least well on the way to getting there. If it's not there yet, it'll be there. By end of year, for certain.

25:00 Alecreally any marketing assets where it makes sense to have somebody on camera talking about something, it now also makes sense, to just use AI to do it

61:00 Aleceverybody's running broad. Your creative's doing the targeting at this point.

Who wrote this

Every AI4NTP post is written by an operator who was in the room when the work happened.

Alec Saluga LinkedIn →

Partner at AI4NTP

A former B2B salesman with no technical background who self-taught AI. He builds and deploys AI-driven marketing and websites, and has grown a following of over 15,000 teaching AI adoption.

Justin Novak LinkedIn →

Partner at AI4NTP

Founder and host of AI4NTP. He sold his first company from his college dorm room, and as a fractional CMO has helped scale multiple businesses past $50M in ARR.

Ian Kilpatrick LinkedIn →

Partner at AI4NTP

A designer, developer, and serial entrepreneur writing code since age 10. He has worked with Disney, the Golden Globes, and the AMAs, and now runs a fleet of AI agents doing real work every day.

How to Make UGC Ads With AI (30 Seconds, Not $250)

Key takeaways

What changed about ad creative?

How do you actually make one?

Are AI UGC ads believable yet?

Do you still need audience targeting?

Where else do AI avatars work?

The repeatable order of operations

Generate the product image

Draft the script with an LLM

Assemble the ad in Marketing Studio

Generate, review, regenerate

Run broad and let the data pick

Frequently asked questions

Tools used in this post

Where this came from

Who wrote this

More field notes

Want to watch this happen live?