Creating a photorealistic AI avatar used to require recording yourself or uploading a selfie. That’s no longer necessary. With the right tools and prompts, you can generate a fully photorealistic avatar from scratch and use it to narrate videos, tell stories, or present content.
In this guide, I’ll walk you through the process using AI tools to generate an avatar, uploading it into HeyGen, assigning a voice, and creating a short narrated video. Best of all, the process can be done using free tools.
I’ll also share a few lessons learned along the way so you can avoid the mistakes I made when I first started experimenting.
Step 1: Start With a Clear Plan
The goal was simple: create a photorealistic AI avatar that could narrate stories in HeyGen (affiliate link).
Instead of cloning my own face with a selfie, I decided to create a completely original character. The avatar would be an African-American man, in his early 60s, friendly and approachable. I also wanted him to look like a natural storyteller.
Once the avatar was created, the plan was to upload the image into HeyGen and use it as a talking narrator.
Step 2: Use AI to Generate a Detailed Image Prompt
Good images start with good prompts.
When I first began experimenting with AI image generation, my prompts were vague, and of course, so were my images.
This is where using ChatGPT to help build a detailed prompt became valuable. Instead of continuing to come up with vague prompts, ChatGPT filled in all of the necessary specifics. Honestly, half of the things it came up with I had no clue existed, let alone what some of them were.
For example, I asked ChatGPT to create a prompt for an image generator that included:
- Age
- Gender
- Facial features
- Expression
- Lighting
- Camera angle
- Overall tone of the character
The result was a much richer prompt describing a photorealistic portrait of a kind older man with warm brown skin, salt-and-pepper hair, a neatly groomed mustache, and a friendly expression suitable for storytelling.
The prompt even included an optional storytelling detail, describing the character as someone who feels like a wise grandfather or community elder. I never would have included those details in my prompt.
Step 3: Generate Avatar Image in Google Gemini
Once the prompt was ready, the next step was creating the actual image. So I clicked over to Google Gemini. I used Nano Banana, the image-creation component of Gemini. Within a few seconds, Gemini generated a photorealistic portrait based on the prompt.
The results were surprisingly good. After a quick review, I selected one of the images to use as the avatar. At this point, all I had to do was download it so I could use it in HeyGen
Step 4: Upload Photorealistic AI Avatar Into HeyGen
With the avatar image downloaded, the next step was to bring it into HeyGen.
Inside HeyGen, I chose to create a new avatar, uploaded the photo, confirmed the platform’s requirements (the first being that I was over 18), and then I named the avatar.
For this example, I named the character Male Narrator.
HeyGen also asked for a few attributes, like gender and ethnicity, to help the system match appropriate voices and expressions.
Once the upload was complete, the avatar was ready for the next step.
Step 5: Choose a Voice for Your Avatar
Without a voice, an avatar is just an avatar. I wanted to create a talking avatar so, I needed to give him a voice.
HeyGen has several options. After testing a few of them, I chose one called Spuds Oxley. His voice was rich, friendly, and slightly playful.
Note: Before generating the talking avatar video, double-check to make sure the correct voice is assigned to the avatar. In my case, the system defaulted to the generic voice, so I had to change it before I clicked Generate.
Once the voice was selected, the avatar was ready to speak.
Step 6: Generate a Script for the Avatar
Instead of writing a script from scratch, I went back to ChatGPT and asked it to create a short story the avatar could narrate.
I asked for a 20-second, family-friendly, comical, and entertaining script. Within seconds, ChatGPT produced a short story about a man attempting to make the world’s greatest pancake.
If you want to see the generated video, skip to the timestamp 8:45 in the YouTube video below.
Step 7: Generate the Avatar Video
With the script ready, the final step was easy. I went back to HeyGen, chose Spuds Oxley (my male avatar), pasted the script into the interface, and selected “Generate”.
After a short processing time, the finished clip was ready. Spuds delivered the story naturally. The lip synching and movements were natural and believable.
It’s a surprisingly simple workflow once you see the steps.
What I Would Do Differently
When I first started experimenting with photorealistic AI avatars, I used ChatGPT to generate the images as well. However, later, when Google released its Gemini Nano Banana image generator, I started using it more often because I liked Nano’s images better.
That got me to thinking, maybe a better workflow might be to ask Google Gemini to create a prompt for its own image generator. It would save me from going back and forth between ChatGPT and Google Gemini.
Both tools are capable, so I decided to experiment.
A Fun Experiment With the Grandkids
To test how different AI tools interpret prompts, I ran a simple experiment with my grandkids.
We used the same prompt in the two AI platforms
The first results didn’t quite look like my grandson. I used the results as a teaching moment. The lesson was two-fold. First, details matter. The second was AI bias. If you don’t specify skin color in your avatar creation, it will not default to brown skin. When we added the skin tone to the prompt, the result looked much closer to him.
Having watched what happened with her brother’s prompts/results, my granddaughter made sure to include skin color from the start. Her prompts produced more accurate images right away.
It was a great reminder that the quality of AI output usually depends on the quality of the instructions you give it (and AI’s bias).
The ABCs of AI Avatar Creation
From my experience, I summarize the process in three steps.
A. Ask AI for a strong prompt. Use ChatGPT to help create a detailed image description.
B. Build the avatar image. Generate the photorealistic portrait in an AI image generator like Gemini.
C. Create the talking avatar. Upload the image into HeyGen, select a voice, and add a script.
Once you understand these basics, creating avatars becomes straightforward.
Try It Yourself
You don’t need expensive software or advanced tech skills to create talking AI avatars. With tools like ChatGPT, Gemini, and HeyGen, you can generate characters, voices, and narrated videos in just a few minutes.
The more you practice prompting, the better your results will become.
Now that you know the basics, why not give it a try? As I mentioned earlier, you can try this for free. So, what are you waiting for? Go out there and get to prompting!