Generating Conversation Videos with InfiniteTalk

InfiniteTalk is an image-to-video and video-to-video AI tool that can be used to generate videos of talking avatars for conversations, explanations, or scenarios. I tested it out to create some dialogue of a phone call between two characters.

One of the big differences in InfiniteTalk’s model is that you can generate much longer clips than the typical short clips created in most AI video tools. In fact, they say you can “generate videos of infinite length. This means users are no longer limited to 10 or 15 seconds – they can create content lasting minutes or even longer, as long as their computer has sufficient RAM and VRAM to handle the processing.”

InfiniteTalk is open source; you can install and run it locally if your computer can handle it. However, I didn’t want the hassle of installing it. I used WaveSpeed’s interface for InfiniteTalk instead.

Watch the video

If the video isn’t embedded below, you can watch it on YouTube. While there are a few glitches and some exaggerated movements, I think InfiniteTalk did well overall at transforming the images and audio into video.

I adapted the script from dialogue in my blog post Selling Storytelling in Learning.

Video generation process

InfiniteTalk generates video based on audio. You give it an image as a starting frame and an audio file, and then it combines those elements to generate the video.

Here’s the process I used:

1. Generate images of the two characters in Flux.
2. Generate audio with ElevenLabs.
3. Generate all the clips for one character using InfiniteTalk (via WaveSpeed).
4. Generate all the clips for the second character using InfiniteTalk.
5. Combine everything in Camtasia.
6. Upscale the video using WaveSpeed’s AI Video Upscaler.

This took me around 2 hours to generate and assemble everything. The cost to generate all of the video clips was about $2.40. Upscaling the video cost more than the actual generation; that cost $3.42. So, for under $6, I was able to generate almost 2 minutes of video.

This workflow was a little clunky though. It was fine for experimenting and creating a sample for my blog, but I wouldn’t want to do a large number of videos this way like for an interactive video. To scale up, I would want a better workflow for generating the audio and video in batches rather than one-by-one. That would probably require a node-based interface like ComfyUI. However, as a starting point, this feels very promising. It’s great to have another option for generating videos.

More on AI

Compare this sample to my experiment with Hedra, another tool for generating videos. Check out this collection of my posts on AI where I “show my work” as I test out different tools and share what I learn.

---

Upcoming events

Instructional Design Uncovered: A Community Conversation with Instructional Designers. September 10. ATD Research Triangle Area

Creating High-Quality AI Media: Prompting, Scripts, and Consistency. October 8. Register for the ATD AI Intensive.

Creative Solutions with Twine: Design, Streamline, and Refine Branching Scenarios. October 14. ATD Dallas Learning Technology SIG

DevLearn

Register for DevLearn by October 2 for the Advanced Rate. Save an extra $100 with code CHRISTY.

Branch Smarter, Not Harder: Naming Conventions, Variables, and Bottlenecks in Storyline

• November 3, 2026 at 9:20 AM
• CE01: Storyline & Rise Pro Lab

Cohesive by Design: Generate Consistent AI Icons, Illustrations & Characters

• November 4, 2026 at 10:00 AM
• BYOD (Bring Your Own Device)

Buzz Discussion: Video & Media

• November 4, 2026 at 3:45 PM