/video — Create Videos

Generate videos from text, images or clips The /video command creates AI generated video from a prompt. Type a scene description, animate an uploaded image, or hand a model an existing clip to edit...

Generate videos from text, images or clips

The /video command creates AI-generated video from a prompt. Type a scene description, animate an uploaded image, or hand a model an existing clip to edit or extend. The picker groups 16 model families — budget drafts to cinematic flagships — with Fast, Standard and Pro quality rungs where a family offers them.

How to use

/video Ocean waves crashing on a rocky shore at golden hour

The /video command with model selector and prompt

Or attach an image and use /video to animate it. Attach a video instead for modes that transform or continue existing footage.

The options panel sets the model family, version, mode, duration, aspect ratio, resolution and audio (where the model offers a toggle) and shows the exact credit estimate before you run.

/video requires a Pro, Ultra or Max plan, or a team seat. See plans.

Models

Rates are credits per second of output at the family’s base resolution, before resolution, audio or Pro multipliers.

Family	From	Duration	Notes
Kandinsky 5	4 cr/s	5 s	Cheapest drafts; silent
PixVerse (C1, V6)	6.25 cr/s	1–15 s	Optional audio; V6 adds Transition and Extend modes
Veo 3.1 (Lite / Standard / Pro)	7.5 cr/s	4–8 s	Google flagship; native audio; Pro rung 50 cr/s
HeyGen (and v5 Digital Twin)	8.5 cr/s	up to 60 s	AI presenter — avatar, voice and scene; billed on delivered seconds
Vidu Q3 (and Turbo)	8.75 cr/s	1–16 s	Native audio; subject-consistency Reference mode; Turbo animates images fast
Grok Imagine (and Video 1.5)	12.5 cr/s	2–15 s	Audio included; Transform mode edits clips
LTX (2.5 and 2.3)	15 cr/s	6–20 s	Up to 4K; Fast and Pro rungs; auto-duration option
Flux 3 (and Draft)	15 cr/s	5–20 s	Black Forest Labs; native audio; widest ratio support
Kling (O3, 3.0, 2.6, O1)	17.5 cr/s	3–15 s	@Element and @Image references; Transform, Continue and Animate Frames modes; Motion Control up to 30 s
MiniMax (H3 and Hailuo 2.3)	20 cr/s	5–15 s	Up to 4K; first five reference images free
Luma Ray 3.2	25 cr/s	5 or 10 s	Cinematic look; silent
Sora 2	25 cr/s	4–12 s	OpenAI’s flagship; silent
Wan (2.7 and 2.6)	25 cr/s	2–15 s	Multi-shot clips with background music
Gemini Omni Flash	32 cr/s	3–10 s	Synchronized audio
Happy Horse (1.0 and 1.1)	35 cr/s	3–15 s	Up to nine character references; video Edit mode
Seedance (2.0 and 1.5)	60.5 cr/s	4–15 s	ByteDance flagship; references can mix image, video and audio mentions; Pro rung 75.9 cr/s

Aspect ratios and resolution

16:9 — widescreen
9:16 — vertical (TikTok, Reels, Shorts)
1:1 — square
4:3 / 3:4 — classic and portrait
21:9 — ultrawide cinematic (Flux 3, PixVerse, Luma, MiniMax, Happy Horse, Seedance)
Model-specific extras include 2:1, 3:2, 2:3, 5:4, 4:5 and 9:21; Flux 3 also offers auto.

Resolutions run from 360p drafts to 4K (LTX, MiniMax H3, Kling 3.0). Higher resolutions multiply the per-second rate — LTX 2.5 at 4K costs about 3.3x its 720p rate.

Audio

Many models generate native audio (Veo 3.1, Flux 3, Vidu Q3, Kling, Seedance, Happy Horse, PixVerse, Gemini Omni, LTX). On some it is included in the rate; on others an audio toggle changes the per-second price — Kling 2.6 and Veo 3.1 Pro roughly double with audio on. Sora 2 and Luma Ray 3.2 generate silent video.

Beyond text-to-video

Image-to-video — attach an image to animate it; supported by most models.
Reference images — keep characters and objects consistent: MiniMax H3 (first five free), Happy Horse (up to 9), Grok Video 1.5 (up to 7), Kling O3 (@Element / @Image mentions), Vidu Q3 and Seedance 2.0 (image, video and audio mentions).
Edit or transform — natural-language edits on an uploaded clip: Grok Imagine Transform, Kling O3 Transform, Happy Horse Edit.
Extend / continue — generate the next shot from a clip: PixVerse V6 Extend, Kling O3 Continue.
Start/end frames — Kling O3 Animate Frames interpolates between two images.
Motion control — Kling 2.6 applies the motion of a reference video to a character image (up to 30 s of character motion).
AI presenter — HeyGen builds avatar, voice and scene from the prompt alone, billed by delivered seconds up to 60 s.

Modes billed on the uploaded clip’s length accept inputs up to 60 seconds.

How billing works

You pay per generated second: model rate × duration, multiplied by resolution, audio and Pro-tier factors, plus small surcharges for extra reference images on some models.
Base rates span 4–76 credits per second across the catalog.
The estimate is shown before you run. Credits are reserved up front, the final cost is settled to the actual charge when the video completes, and failed generations are refunded automatically.
Videos land in the chat. Save them to Keep for 1 credit or download the MP4.
Upscale a finished clip with /upscale, billed per output megapixel.

Tips for great videos

Describe motion explicitly — “a hummingbird hovering near a red flower, wings blurred” beats “a bird”.
Keep prompts focused — one subject, one action, one camera move.
Anchor with a reference — an uploaded start image or character references give far more predictable results than text alone.
Draft cheap, finish strong — iterate at low resolution on a fast model, then re-run the winning prompt on a Pro rung.

Related: /image, /upscale, /sound

More in Slash Commands

See Krater pricing or contact support.