Paula is our AI employee Dentro.
She is our AI influencer.
This video for YT shorts and TikTok took us 4 hours to make:
It was our first attempt though.
I’ll show you how to make AI influencer videos in the same time or less.
Challenge
It’s difficult to create long form videos of Paula that are lip-synced.
Because creating moving avatars that are character consistent…
- is rather expensive (0.6$ per 5 second clip)
- takes a lot of time (1 – 3 min)
Idea
Creating Paula’s voice only is much faster and cheaper than creating a lip-synced video 🤔
- Create only a short intro video with Paula and her face 🪞
- The rest of the video uses Paula’s voice, but not her face 🗣
For any short form video, we also need to decide on a single concise topic.
Our company focuses on AI.
So Paula explains a simple but powerful way to make your LLM apps more reliable!
Let’s dive into the exact steps how to make AI influencer videos…
Steps to create the INTRO video
- Create a still image of Paula
- Use the image to create a video of Paula
- Create the audio of what Paula says
- Lip-sync the audio with the video
1. Create a still image of Paula
- Use Midjourney with character consistency 🫱🏽🫲🏾
- We pay the 10$ plan for Midjourney
- Caveat: your images are public if you only pay the 10$ plan and not the 60$ plan 🤦🏽♀️
- I explain more about this in my comparison video about Midjourney and Ideogram
2. Use the still image to create a video of Paula
Use Chinese Minimax-Video-01 for this via Replicate 🎥
- Upload the still image
- Give it a prompt (e.g. “a woman explaining something”)
- Wait for the video 🎉
This costs ~0.5$ per video.
Replicate is btw one of the 3 Sexy AI API Keys to Access More Models than Hugh Hefner
3. Create the audio of what Paula says
Use Elevenlabs for creating the audio:
- Create a custom voice for Paula that can be re-used later on
- I recorded my own speech with what I want Paula to say for the intro video
- Use the elevenlabs voice changer to change my voice to hers!
Why Elevenlabs?
There are many text2voice AI models out there. The most famous and known voices are probably the voices of ChatGPT. ChatGPT Plus users can talk with the AI.
1️⃣
These voices however, can’t be customized.
Paula is part of a team of Austrians.
Therefore we want her to have Austrian accent.
With Elevenlabs custom voices we can do this!
2️⃣
Another issue with text2voice models is that the generated speech can sound unnatural.
The AI doesn’t pause at the right moments.
The AI doesn’t emphasize words as a human would.
With Elevenlabs voice changer we can achieve this!
First I record my own voice with the right pauses and emphasis on words.
Then we simply change it to Paulas voice!
4. Lip-sync the audio with the video:
- we need to make sure that Paula’s mouth in the video moves along with her audio
- Use Sync for this
- It’s cheaper than RunwayML for lip sync. Although the quality for certain videos is better via Runway imo.
- Costs another ~0.1$ for the 5 second clip
➡️ we finally got the intro video! 🎉
Steps to create the REST of the video
- Use again Elevenlabs voice changer for the audio.
I just recorded everything that needs to be said and changed it to Paula’s voice. - Record my screen to show websites and the python code using langchain.
We use Tella for this. - Create more videos with Minimax on Replicate.
Just seeing code might be boring for generation TikTok.
That’s why we added the video of the man angry with his laptop 😆 - Create an explainer image via Replicate (the tight rope walker).
Use Canva to add the Logos of OpenAI and Anthropic.
This image should nail down the main point of the video. - Use CapCut to put it all together.
Videos, Audios and images.
We also added some effects.
Conclusion
Right now the process is still pretty involved.
A single human is probably still faster to create such short video.
However, we’ve come far:
- GenAI is responsible for large parts of the video
- Anybody can use this process, both technical and non-technical people 🤟🏽
Now you know how to make AI influencer videos!
Follow Paula on Social media to stay up to date on what’s possible and make AI influencer:
