Dec 13, 2024 4 min read Tutorial

Getting Started: How to use Infinity AI

There are several text-to-video or image-to-video AI models available (Luma, Runway, Pika, etc) but not many that allow your characters to speak.

People are at the center of stories. And if you have expressive, talking characters, you can tell good stories even if you don’t have anything else.

To that end, we launched the Infinity V2 model, which allows you to animate ✨infinite✨ actors. See example user videos in the Infinity Gallery.

This is a step-by-step tutorial showing how to use Infinity to generate AI talking head videos.

0:00

/0:49

1 min tutorial on how to use Infinity (made using an Infinity avatar!)

Step 0. Navigate to the Infinity Studio

Infinity V2 is a video foundation model focused on expressive characters that talk: infinity.ai/studio. Login to get access to the full set of features.

Step 1. Generate an image

The first step is to pick the image of your character. Any style of image - photos, paintings, 3D cartoons, etc - works with Infinity. Please note that animals and non-humanoid characters (e.g. a toaster with a face) do not work very well at the moment (though you can still make it work with a lot of persistence). There are 3 options for the image:

Tip: Infinity works with input images of ANY aspect ratio. Make sure "Crop face" is OFF so that the image's original aspect ratio is used.

Option 1. Select a character from the library

Option 2. Generate a character using the text-to-image tool

Option 3. Upload an image file

Step 2. Generate audio

Next, we need to define the audio that we want our character to say. Similarly, there are a few options for the audio:

Option 1. Type out the script

Go to the “text-to-speech” tab and write out the exact script you want your character to say. Select a voice from the dropdown menu and then press “generate audio.”

Tip: Use punctuation to control the delivery of your script. “-” and “...” for pauses. “!” and CAPITALIZATION for emphasis. Each time you press “generate audio” the audio will be different.

Option 2. Record your own voice

Go to the "record" tab of the audio section and record yourself directly.

Option 3. Upload an audio file

Upload any audio file with talking or singing in it. Note: long sections of silence or instrumentals will lead to weird behavior. Use active speaking/singing audio only.

Step 3. Generate the video

Finally... generate your video! Here’s an explanation of the generation parameters:

Resolution: 100k, 250k, 400k (experimental). These refer to the TOTAL number of pixels in a single frame. For example, if your video aspect ratio is 1:1 square, then a “250k resolution video” will be 512 x 512px. If it’s 16:9, then a “250k resolution video” will be 672 x 384px.
Crop Face: ON or OFF. If OFF, then your entire image will be uploaded. If ON, then your image will be cropped to a square around the character’s face.
Stability Mode: Expressive, Medium, Stable.
Num Videos: 1-3. Number of videos to generate at once with these generation parameters.

Our recommendation is to use: 250k resolution, Crop Face OFF, Expressive (or Medium)

Tip: Every time you generate a video, it will be different. Start with the recommended generation parameters and then adjust accordingly. Increase stability if you’re getting a lot of blurring. Increase resolution if the video looks good but you just want more details.

See gallery with more examples: https://infinity.ai/gallery

Additional Studio Tips [advanced]

Save a character

Saving a character (i.e. an image + voice combination) makes it easy to re-use that person in the future. To save a character from a clip you’ve already generated just click the “Replay” icon and then “Save character.” This character will now appear permanently in the Character Library.

Your characters are only visible to you. You can organize and edit your characters by going to the “My characters” page.

Replay

Replay makes it easy to generate the same video clip again. Press the “replay” icon and then “Replay.” This will load the same image, audio, and video generation parameters into the studio.