How do I make an AI character video?

Make one reference image of the character, give it to a video model that accepts image references, and write a short prompt that says what happens second by second, with any dialogue in quotes. A 5-second 720p Seedance 2.0 Fast clip costs about $0.60 on BytePlus^[1] and $0.75 on Replicate^[2], and the character stays recognisable as long as every video starts from the same reference.

What you need

• A reference image. One clear, full-body picture of the character on a plain background, plus a written description you reuse every time. You can draw it, or have an image model design it; see can AI design an original character.
• A video model that takes references. Seedance 2.0 accepts up to 9 images per generation^[3] and Seedance 2.5 up to 30^[4]. Kling 4.0's Omni Reference takes up to 15 assets, including several views of one character^[5]. Veo 3.1 takes up to three reference images^[6].
• A prompt written as a timeline (below).
• An editor for trimming, music and captions. Ask the model for no on-screen text and add your own.

Do I need a reference image?

For one-off clips, no: a detailed text description works. For a character that has to look the same in the next video, yes. In our own work, clips that start from a painted first frame or a reference image keep the face far better than clips described in text alone. Our explainer character Penny was built from 31 separate clips, each starting from a frame painted off one master image, and reads as one character across 82 seconds. The method is in keeping an AI character consistent.

How to write the prompt

In our renders, Seedance followed time ranges far better than prose. Use one beat per few seconds, one action per beat, and exact dialogue in quotes:

References:
@Image1: the character. Keep her face, hair and outfit exactly.

Visual style: photoreal phone footage, warm evening light.

Avoid: No subtitles. No on-screen text. No change to her face.

One continuous handheld take, eye level.
0-3s: She bursts into the kitchen holding a burnt cake, smoke curling off it.
3-7s: She sets it down and says: "It's fine. It's caramelised."
7-10s: A piece falls off with a clunk; she stares at the camera, deadpan.
Sound: smoke alarm chirp, a plate clatter.

If you would rather type one line and let a tool expand it, make an AI character video from text shows that, with a real before-and-after. Our free video prompt builder writes the timeline format from a few picks, and the prompt library has the full prompts behind our published videos.

What it costs and how long it takes

Render time varies by provider and queue, and we could not verify published render times for the tools above. In Cast a video usually lands in about a minute. Budget for retakes: our rule is one take at a time, changing the prompt between takes, because batches of spare takes double the bill. For a full price breakdown, see how much AI video costs per second.

Common mistakes

• Too many events. Five seconds holds about one beat; ten seconds, two.
• A face that looks like a celebrity or a franchise character. Video models block these after the render. Design an original face.
• Leaving the outfit to the model. Write it out every time, or it drifts.
• Asking for a phone-camera look without saying the camera is invisible. In our renders, the model sometimes painted a phone into the shot until we wrote that the camera itself is never visible.

Where Cast fits

• You want the whole workflow in one phone app: pick a free Malfoid & Friends character or describe your own, type one line, and get a 9:16 clip of 5, 10 and 15 seconds.
• You want the price shown before you render (52 credits for 5 seconds on Seedance 2.0 Fast) and failed renders refunded automatically.
• You want up to three characters in one video, each kept on-model from its own reference.

Where it does not:

• Starting from your own image; Cast characters are drawn from a text description.
• Clips longer than 15 seconds, edited sequences or multi-episode stories (Katechon Arc does those: https://arc.katechon.technology/).
• Full manual control over every shot, or lip sync to your own audio.

People also ask

What tools do I need to make an AI character video?

An image model for the reference, a video model that accepts image references, and a basic editor. Some apps combine the first two.

How long does an AI character video take to make?

In Cast a render usually lands in about a minute; other tools vary with queue, length and resolution. Designing the character and fixing the prompt take longer the first time and almost no time after.

How much does a character video cost?

At API rates, a 10-second 720p Seedance 2.0 Fast clip is about $1.20 to $1.50^[1][2]. In Cast, 10 seconds costs 99 credits, about $1.10 at the Starter rate.

Can I make a character video for free?

For a few clips, yes. See free AI character video generators for current free tiers.

Sources

Claims about other companies come from their own pages, read on the date shown. Products and prices change; check the link for the current version.

1. BytePlus ModelArk: model pricing (read 2026-10-18)
2. Replicate: bytedance/seedance-2.0-fast (read 2026-10-18)
3. ByteDance Seed: Seedance 2.0 launch (read 2026-10-18)
4. ByteDance Seed: introducing Seedance 2.5 (read 2026-10-18)
5. Kling AI: AI video generator (read 2026-10-18)
6. Google AI for Developers: Veo (read 2026-10-18)
7. Google AI for Developers: Gemini API pricing (read 2026-10-18)
8. Runway help: how credits work (read 2026-10-18)
9. Runway: pricing (read 2026-10-18)

Related answers

• How do I keep an AI character consistent across videos?
• Can AI design an original character for me?
• How do I write a prompt for an AI video?
• Is there a free AI character video generator?
• Do I need a talking avatar or an AI character video?
• How much does AI video cost per second in 2026?

Part of the Cast answers library (Your own characters, OCs and fan characters).

Updated 2026-10-18