Hedra Character 3 Video Generation

Hedra is one of the image to video tools that is particularly good at syncing to audio. There are plenty of AI video tools that can show people walking around or dancing, but not as many that do lip syncing for realistic conversations. Since I had tested Hedra last year, I wanted to experiment with it again to see how much Hedra Character 3 has improved over the earlier versions. I tested the same character with different emotions.

Hedra Character 3 Video Generation. A woman sitting at a sleek white table in a modern office.

Hedra character video example

Here’s my example. This is two clips: one with a calm audio and one showing the same character frustrated. (Email readers, if the video doesn’t appear below, try watching it on YouTube.)

How I created this

I’m using the free plan for Hedra currently, so I wanted to save my credits just for the video generation. Therefore, I generated my character image and audio outside of Hedra. You could do it all within this single tool though.

First, I generated the character image in Midjourney. You need an image with a clear face to generate a video. I specifically prompted for glasses in my character image because sometimes those throw off the tools and generate weird artifacts.

Editorial photo, Latina woman wearing a red button-down shirt and wire-rimmed glasses, sitting in a modern office conference room, looking at the camera, neutral expression

AI-generated photorealistic image. Latina woman wearing a red button-down shirt and wire-rimmed glasses, sitting in a modern office conference room, looking at the camera, neutral expression.

I also generated my audio separately in ElevenLabs. Again, this was to save my credits in Hedra during my experiment; it would be simpler to just generate audio directly in Hedra.

From there, it’s easy to generate a video. With external assets like this, I just uploaded my image as the starting frame and my audio in the “audio script” section. In the second clip, I added a text prompt of “annoyed, frustrated” to set the emotion.

On a free plan, you’re limited to a max of 20 seconds per generation. That’s enough for a few sentences of dialogue.

Hedra for scenarios

These snippets of dialogue are from my New Hire with Attitude branching scenario. (That’s why there’s two different versions of similar information; those are the consequences of different decisions you make as you try to resolve the conflict.)

I think you could use this to create multiple snippets and show different versions in a scenario. The videos still have flaws, but the hand gestures felt pretty natural (if a little too frequent). Even the reflections on the glasses turned out well. This would be good enough for generating dialogue for an interactive video scenario, at least for some purposes.