Loading video...
Video Failed to Load
Your AI UGC looks fake cause you're missing this extra step.
233,451 views • 23 days ago •via X (Twitter)
82 Comments

Bro we agreed you won’t share all the sauce for free 😭😭😭

But we already have too much 😭

This guy is so handsome and smart

W Value

Learning from the bests in the game - this team is cracked!! Thank you man 👊

someone else would’ve sold this as a $1999 course. 🐐

We gatekeep nothing! Appreciate you man 👊

what'd you use to film this video? looks rad

Just my Iphone 14 and @TellaHQ 😎

What a great vid! Thanks. Not often do you see someone advocating an older ai model 😁

@pounddz 100% agree. For AI UGC in particular, you are not aiming for perfection. NBP was trained on real Iphone shots and in combination with Gemini Omni, remains undefeated for AI UGC but ONLY when combining it with a real reference image

So much sauce in a single video

Appreciate you! Thank you 👊

too much value!

Yoo thank you man!

you're g

Thanks bro! What's your AI UGC workflow right now? You're doing DTC right?

🐐🔥

👊

Wen .md file

Ooo not a bad idea 👀

🔥🔥 Value

Thank you man! Big praise from the Starpop founder himself! Let me know how you get on trying out the new workflow!

Will do!

missing that extra step can definitely make or break authenticity in UGC

100% agree. That Waxy effect on the skin is the ultimate giveaway for AI UGC

Dude is actually so cracked !

Crazy that such a small step can have such a big change in the output!

People need to pay more attention to this .

What is your opinion on using seedance 2.5?

I actually don’t rate Seedance, atleast, not as high as everyone else does. For AI UGC, there is nothing that comes close to Omni right now Imo and even for AI Animation, Minimax has been crushing it for me recently. Better outputs and a fraction of the cost. The one advantage Seedance currently has is 30 second durations but if you’re a bit of a control freak with your ads and want to make sure every detail is dialed, generating a 30 second clip is never going to hit the mark. It’s a speed advantage but at the cost of quality

super helpful, cheers mate.

For sure man, appreciate you! 👊

VERYYYYY VALUABLE 👾✨🤝🎀💗🐐

Let’s gooooo!

Why did I watch this whole thing

Haha hope that was half an hour well spent!

Thank you bro i’ve been tweaking out w seedance recently, def trying this tomorrow. how much does one of these cost you roughly?

For sure man! I don’t actually rate Seedance, especially not for AI UGC. Like anything, depends on the length of the video etc. Should be cheaper than Seedance I would imagine. I’m testing a Minimax flow this week that if it performs, will cut costs dramatically

would love a video on storyboarding!

Sick idea, I want to film a part 2 for this to address people's questions and then I'll get right on that! Thanks for the feedback, really appreciate it!

Sauce

🧑🍳

watched this video just now and completed your 1k follower haha🤣🤣 GOATED stuff man!

Ayoooo, let's go! Thank you man, glad you found it useful! Part 2 out later today 💪

curious what the extra step is tbh

I’ll put out an MD file later today but there is more sauce in that video than that for sure! The extra step though is getting a real reference image taken from a real UGC creator with real noise and imperfection baked into the reference that the human eye can’t see (but exists as data). Without having this as your base, the video model doesn’t have that data to animate and instead fills that gap with empty data, leading to that waxy AI effect you get on the skin, you know?

the waxy skin thing checks out, that gap-fill is always the tell

Crazy value man thank you

Appreciate you man! This workflow has been cooking!

The image stuff is deep man. Really went in depth on it

That's the secret. You gotta spend way more time on your reference images than on the image generation. Lay the foundations first

100% agree

Good sauce 🔥

Appreciate you 👊

sauce man

Thanks man, glad you found it useful 👊

Which step is it though?

Using a real tiktok frame as a reference image with real Iphone noise baked into the photo. If you use an AI generated photo as your base, even if it looks real to the human eye, when you animate it, it will have that waxy AI shine baked into it

Waxy shine is such a giveaway.

100% agree! Before, no one knew how to get rid of it, now that we’ve figured it out, everyone’s AI UGC should get to the next level 💪

Real noise beats fake polish.

How do you make voice stay consistent throughout multiple generations?

Locking the prompt. That's why we spend so long on the first generation. Once that's dialed, the only thing you change is the dialogue, everything else stays locked which results in a consistent output

Got it. One more thing, you mentioned using Google Omni Video, not Omni 1.1 Flash, as far as I'm aware both use the same Omni 1.1 Flash structure, what's the actual difference between these two? Am I missing something? Why do you prefer the first one?

Either work man, the real unlock is the IRL reference image and just making sure you've got that digital data that the human eye can't see baked into the reference. As with anything, don't fixate on the model too much, these things will change and they'll change fast!

Thanks a ton!!

bro is him!

g

sauce

chef

Way too many sauces, this is the best video on creating AI UGCs, thank you Kristian

HUGE praise from the GOAT of AI ads! Thank you so much Narek, appreciate you massively man! 👊

@pounddz Goated video

@pounddz Yooo thank you!

chef

g

Thank you for this!! This is amazing!! I am already in the community, also added you on discord as friend (Scared to text cus I don't wanna disturb you), but wanted to let you know that you are a star!!

Yooo, appreciate you man 👊 ET goated fr

Thanks for this man.

Yessir! Hope it helps you print mate 🤑

I am very sure it will I will keep in touch!
