MitchC
Silver Contributor
FASTLANE INSIDER
EPIC CONTRIBUTOR
LEGACY MEMBER
Read Rat-Race Escape!
Read Fastlane!
Read Unscripted!
AI UGC experiment and reviews
If you follow my progress thread, you know I've been running AI voice ads with Elevenlabs and using old B-roll and stock footage to produce ads.
I've been seeing people use AI creators and knew it was possible, so I decided to spend a day trying.
The TLDR is don't bother. The method I've been using, or just using UGC creators, is still the best and most efficient.
But I'll post this here as there were some useful findings, and maybe people can add to and improve it. Plus I can come back to it as a reference later so I didn't completely waste an entire day with this shit.
What did I try to create?
A consistent AI person, who I could make talk in different scenes, which could be cut together to form a narrative.
I didn't try adding the product to these scenes; I didn't get that far.
Process:
Generate AI Voice
Use Elevenlabs for this. The new create your own voice from a prompt thing is amazing, but it's so weird. You save the voice, and it sounds completely different every time you use it. So you need to get it to read the entire script in one shot, or it'll be a different voice every time. Hopefully, they fix this. Or just use the old pre-made voices.
Generate an AI person
I don't think it would be legal to use an existing creator from our ugc, and it would be weird anyway, I wouldn't want someone doing that to me, so step one is to generate a fake person
For this I used:
Dreamina from capcut
It created 4 photos of the same person from a prompt written by ChatGPT. The person matched the prompt and looked ok, looked ai though.
I'm not sure if it was just luck that it generated the same person in different poses, but it did.
Optional step I didn't do is use Enhancer AI to make the AI photos look more like a real person.
The next step is to create this as an avatar you can continue to use, for this I used:
OpenArt
I took the four photos from Dreamina and added them to OpenArt to train it to create an AI avatar of that person.
I then used OpenArt to create photos of that person in different places
It did this well; the person was consistent, but the prompts needed to be good to get the scene right. I would suggest that you use ChatGPT to help.
The next step is to make the photo into a video of the person saying the audio.
Animating the photo worked well, but making the mouth line up with words and look real is where it fell apart.
For this, I tested many different apps.
The best was probably Hedra, but it still was not realistic enough. It's also expensive and for some reason maxes out at 720p lol.
Lipdub is meant to be great, but you need 30 seconds of talking footage minimum to train it. This is impossible if you are trying to use an AI avatar. You can generate 10 seconds of your AI avatar talking from OpenArt or Dreamina, but using this to train Lipdub doesn't give a good result.
I also tested Google Gemini Veo3 and want to review it because it is a piece of shit. I paid $220, but luckily, they refunded me.
Gemini itself was completely useless at helping me, giving me images when I was asking for prompts, giving me big articles with references when I was asking for a simple website link etc. ChatGPT is far far far superior.
So, back to the UGC creation with Veo3.
First of all, I will say it does generate the most realistic talking videos.
However, there are major drawbacks with how it does that.
Apparently, you can get it to create consistent people, but I have no idea how.
So the issue is:
This thing cannot start with an image reference, even though it lets you upload an image. You also can't even upload your own audio at all. It works entirely off text prompts.
The other apps allow you to begin with an image reference and either add your own audio or have the app create it. This is far more useful.
It's also super expensive, very slow, very limited, and everything it generated was horizontal, no matter how much I asked for vertical.
Conclusion:
Generating an AI image, using it to create an AI avatar, using an AI voice, getting the AI to animate the AI person saying the AI voice, results in a video that is clearly AI. No surprises here really. The problem is, if you replace one step of this with a real person, you may as well just use a real person for the entire thing.
Things that could work:
I've seen people using face swap and voice changers to create videos of them as someone else, this could be useful.
Photos and anything other than someone up close talking to a camera. You could maybe zoom out or hide their mouth a bit somehow. You can also add grain to photos and videos or filters so it's not as obvious.
I don't need that right now, but my new brand will, so I plan to test it again with that. The issue there will be whether it can add the product accurately enough. I have used AI to generate images in the past and then just photoshopped the product in, that works well, but I think AI is better since then.
I've seen people create AI models of themselves holding a product, and the AI avatar will then be holding it, but what's the point of that? Just film the video yourself instead of F*cking around trying to get AI to do that.
It's ironic because I think AI is better at and more likely to replace high-end fashion shoots, expensive special effects, cinematic scenes etc, rather than a person shooting an iPhone video talking to the camera selfie style which I was trying to create.
The method I've been using where you use real b-roll with AI voice. Adding clips of a real person talking, even if they are saying completely different words to the AI voice, will look better than an AI trying to say it.
If you follow my progress thread, you know I've been running AI voice ads with Elevenlabs and using old B-roll and stock footage to produce ads.
I've been seeing people use AI creators and knew it was possible, so I decided to spend a day trying.
The TLDR is don't bother. The method I've been using, or just using UGC creators, is still the best and most efficient.
But I'll post this here as there were some useful findings, and maybe people can add to and improve it. Plus I can come back to it as a reference later so I didn't completely waste an entire day with this shit.
What did I try to create?
A consistent AI person, who I could make talk in different scenes, which could be cut together to form a narrative.
I didn't try adding the product to these scenes; I didn't get that far.
Process:
Generate AI Voice
Use Elevenlabs for this. The new create your own voice from a prompt thing is amazing, but it's so weird. You save the voice, and it sounds completely different every time you use it. So you need to get it to read the entire script in one shot, or it'll be a different voice every time. Hopefully, they fix this. Or just use the old pre-made voices.
Generate an AI person
I don't think it would be legal to use an existing creator from our ugc, and it would be weird anyway, I wouldn't want someone doing that to me, so step one is to generate a fake person
For this I used:
Dreamina from capcut
It created 4 photos of the same person from a prompt written by ChatGPT. The person matched the prompt and looked ok, looked ai though.
I'm not sure if it was just luck that it generated the same person in different poses, but it did.
Optional step I didn't do is use Enhancer AI to make the AI photos look more like a real person.
The next step is to create this as an avatar you can continue to use, for this I used:
OpenArt
I took the four photos from Dreamina and added them to OpenArt to train it to create an AI avatar of that person.
I then used OpenArt to create photos of that person in different places
It did this well; the person was consistent, but the prompts needed to be good to get the scene right. I would suggest that you use ChatGPT to help.
The next step is to make the photo into a video of the person saying the audio.
Animating the photo worked well, but making the mouth line up with words and look real is where it fell apart.
For this, I tested many different apps.
The best was probably Hedra, but it still was not realistic enough. It's also expensive and for some reason maxes out at 720p lol.
Lipdub is meant to be great, but you need 30 seconds of talking footage minimum to train it. This is impossible if you are trying to use an AI avatar. You can generate 10 seconds of your AI avatar talking from OpenArt or Dreamina, but using this to train Lipdub doesn't give a good result.
I also tested Google Gemini Veo3 and want to review it because it is a piece of shit. I paid $220, but luckily, they refunded me.
Gemini itself was completely useless at helping me, giving me images when I was asking for prompts, giving me big articles with references when I was asking for a simple website link etc. ChatGPT is far far far superior.
So, back to the UGC creation with Veo3.
First of all, I will say it does generate the most realistic talking videos.
However, there are major drawbacks with how it does that.
Apparently, you can get it to create consistent people, but I have no idea how.
So the issue is:
This thing cannot start with an image reference, even though it lets you upload an image. You also can't even upload your own audio at all. It works entirely off text prompts.
The other apps allow you to begin with an image reference and either add your own audio or have the app create it. This is far more useful.
It's also super expensive, very slow, very limited, and everything it generated was horizontal, no matter how much I asked for vertical.
Conclusion:
Generating an AI image, using it to create an AI avatar, using an AI voice, getting the AI to animate the AI person saying the AI voice, results in a video that is clearly AI. No surprises here really. The problem is, if you replace one step of this with a real person, you may as well just use a real person for the entire thing.
Things that could work:
I've seen people using face swap and voice changers to create videos of them as someone else, this could be useful.
Photos and anything other than someone up close talking to a camera. You could maybe zoom out or hide their mouth a bit somehow. You can also add grain to photos and videos or filters so it's not as obvious.
I don't need that right now, but my new brand will, so I plan to test it again with that. The issue there will be whether it can add the product accurately enough. I have used AI to generate images in the past and then just photoshopped the product in, that works well, but I think AI is better since then.
I've seen people create AI models of themselves holding a product, and the AI avatar will then be holding it, but what's the point of that? Just film the video yourself instead of F*cking around trying to get AI to do that.
It's ironic because I think AI is better at and more likely to replace high-end fashion shoots, expensive special effects, cinematic scenes etc, rather than a person shooting an iPhone video talking to the camera selfie style which I was trying to create.
The method I've been using where you use real b-roll with AI voice. Adding clips of a real person talking, even if they are saying completely different words to the AI voice, will look better than an AI trying to say it.
Dislike ads? Become a Fastlane member:
Subscribe today and surround yourself with winners and millionaire mentors, not those broke friends who only want to drink beer and play video games. :-)
Last edited:
Membership Required: Upgrade to Expose Nearly 1,000,000 Posts
Ready to Unleash the Millionaire Entrepreneur in You?
Become a member of the Fastlane Forum, the private community founded by best-selling author and multi-millionaire entrepreneur MJ DeMarco. Since 2007, MJ DeMarco has poured his heart and soul into the Fastlane Forum, helping entrepreneurs reclaim their time, win their financial freedom, and live their best life.
With more than 39,000 posts packed with insights, strategies, and advice, you’re not just a member—you’re stepping into MJ’s inner-circle, a place where you’ll never be left alone.
Become a member and gain immediate access to...
- Active Community: Ever join a community only to find it DEAD? Not at Fastlane! As you can see from our home page, life-changing content is posted dozens of times daily.
- Exclusive Insights: Direct access to MJ DeMarco’s daily contributions and wisdom.
- Powerful Networking Opportunities: Connect with a diverse group of successful entrepreneurs who can offer mentorship, collaboration, and opportunities.
- Proven Strategies: Learn from the best in the business, with actionable advice and strategies that can accelerate your success.
"You are the average of the five people you surround yourself with the most..."
Who are you surrounding yourself with? Surround yourself with millionaire success. Join Fastlane today!
Join Today
