The hardest problem in AI video is making the same character show up twice. Generate two clips from text alone and you'll get two strangers who vaguely resemble each other. PiTube's Consistent Character feature solves this: lock in a character once, and every clip you generate afterward keeps the same face, outfit, and style.
What Consistent Character Does
Instead of describing your character from scratch every time, you give PiTube a reference image — a picture of the character you want. The video model uses that image as the anchor for identity: the face, hair, clothing, and overall look come from the image, while your text prompt controls what the character does in the scene.
That split is the whole trick. Image = who. Prompt = what happens.
How to Lock a Character
- Open the Studio and choose the Consistent Character option.
- Upload or pick a reference image of your character.
- Write a motion prompt describing the scene and action.
- Tap Generate and pay with Pi as usual.
Reuse the same reference image for your next clip and you get the same character again — walking through a market this time, or waving at the camera, or standing in the rain. One image, many scenes.
Tips for Good Reference Images
- One character per image. A single subject, clearly the focus of the frame.
- Face visible and well lit. Front-facing or three-quarter view beats a profile or a shadowed face.
- Clean background. A busy background can bleed into the video; simpler is safer.
- Show the outfit you want. Whatever the character wears in the reference is what they'll tend to wear in the clip.
- Sharp beats stylish. A crisp, medium-close shot gives the model more identity to hold onto than a distant or blurry one.
Describing Motion Without Changing the Face
The reference image handles identity, so your prompt should stay out of its way. Describe action, camera, and setting — not appearance:
- Good: "she turns and smiles, then walks toward the camera, city lights behind her"
- Good: "he sips coffee and looks out the window, rain on the glass, slow push-in"
- Risky: "a beautiful woman with green eyes and short blonde hair walks..." — you're re-describing the character, and the model may follow your words instead of the image.
Try It
Pick a character, lock the image, and generate two or three clips with different motion prompts. Watching the same face carry across scenes is the moment AI video stops feeling like a slot machine and starts feeling like filmmaking. New to PiTube? Start with How to Make an AI Video with Pi.