Text-to-Video & Image-to-Video
The two ways to generate video on Xotic AI — and why, with image-to-video, the starting image is the single most important thing.
There are two ways to generate a video, and knowing when to use each is half the battle.
Text-to-video (T2V)#
Most actions are text-to-video: you pick a character and an action, and the model generates a clip from scratch. There's no starting image — the model creates every frame. It's the quick, hands-off path, and it's how the bulk of the V1 and V2 actions work.
Image-to-video (I2V)#
Image-to-video starts from a still image and animates that exact frame. The V3 (Custom) and V4 (Talking) models are I2V-only — they can't run without a starting image. You provide the image, the model brings it to life.
The starting image is everything#
This is the most important thing to understand about I2V:
Put simply, the quality ceiling of an I2V video is the quality of its starting image. A stunning still becomes a stunning video; a mediocre still becomes a mediocre video, no matter how good the animation is.
Get the starting image right first#
So the workflow is: perfect the image before you animate it.
- Generate or choose your starting image in the image studio or from a chat.
- Regenerate until it's genuinely great — clean anatomy, the pose and framing you want, the right look. Don't settle. (Use the prompting guide to nail it.)
- Only then take it into I2V and animate.
Images with a clear subject, good framing, and clean composition animate best.
The advantage of I2V#
Because you hand the model the exact frame, I2V gives you total control over the look — the precise pose, outfit, framing, and a face that's exactly your character. T2V leaves those to chance on every generation; I2V locks them in. When you know exactly what you want the video to look like, start from an image.
How to provide the image#
Pick an image-to-video action — V3 (Custom) or V4 (Talking). If you haven't set a starting image, it shows a "Requires image" overlay and the studio sends you to pick or generate one first. Once an image is attached, generate as normal.
For V4 (Talking), you also provide a line of text (up to ~400 characters) — she'll speak it on camera, lip-synced to the words.