Which video model should I use?

The model menu on Video lists the models you can use now, in three parts: Make a clip, From references only, and Work on a video. Each row's badges say what the model takes, such as "Start & end frame", "9 image refs", "Storyboard" or "Makes sound", and each shows its price at the length and resolution you've set. The Takes filters at the top keep the models that take what you have, and hovering over a model shows a line on what it's for.

Models can be switched off for a while, so your menu may list fewer than this page does.

Pick by what you need

• A clip from words, frames and references together. Seedance 2.0, the default, takes a start and end frame, up to 9 image refs, 3 video refs and 3 audio refs, and makes sound. Kling 4.0 makes clips of up to 30 seconds one of two ways, picked by the bar's tab: Keyframes on Text / Image to Video, a start and an end frame with up to 8 frames between them, or Omni on Reference to Video, up to 10 image refs and 5 video refs. It takes no audio refs for now.
• Kling 4.0 for less. Kling 4.0 Flash is its faster, cheaper tier: 3 to 20 seconds at 720p, with sound, the same two ways, Keyframes from words or a start frame (no end frame, no frames between) and Omni from up to 10 image refs and 5 video refs.
• One long take. Seedance 2.5 makes clips of up to 30 seconds, and so do Kling 4.0, Wan 3.0, Wan 3.0 Prime and Grok Imagine Video.
• The most references. Seedance 2.5 takes up to 30 image refs, 10 video refs and 10 audio refs. Wan 3.0 and Wan 3.0 Prime take up to 10, 5 and 5, and a sound on its own.
• A ready-made effect. PixVerse v6 has 49 effects, from a product turned into a landmark to a dance trend, each made from one or two of your pictures at its own length. See How to make a video.
• A clip from a web page or a document. Wan 3.0 and Wan 3.0 Prime make a clip from what a public web page or a document says: paste the page's link, or add a pdf, Word, PowerPoint, Excel, txt or md file. See How to make a video.
• A still to animate. Any model badged with a start frame. Vidu Q4 animates a still or makes a clip from up to 15 image refs and 3 voices, 3 to 16 seconds, up to 4K, with sound.
• A fast clip from words or a still. Kling 3.0 Turbo makes 3 to 15 seconds at 720p or 1080p. PixVerse v6 makes 1 to 15 seconds from words, a start frame, or a start and an end frame, from 360p to 1080p, with sound if you switch it on. PixVerse v6 Max is its finer tier, made for ads and short drama, from words or a start frame, at a little more. Grok Imagine Video makes 6 to 30 seconds from words or a start frame, from 480p to 1080p. Grok Imagine 1.5 makes 1 to 15 seconds from words, a start and an end frame, or up to 7 image refs, one at 1080p, with sound unless you switch it off.
• Several shots in one clip. Kling 3.0's Storyboard: up to 5 shots, each with its own prompt and length, 15 seconds in all. Kling O3 (reference) takes up to 6, each up to 12 seconds and 15 in all, made from your image refs.
• Sound made with the picture. The models badged Makes sound: the Seedance models, Kling 3.0, Kling 4.0, Kling 4.0 Flash, Kling O3 (reference), Vidu Q4, Wan 3.0, Wan 3.0 Prime, MiniMax H3, Grok Imagine 1.5, PixVerse v6 and PixVerse v6 Max. See Sound in your video: voices and lip-sync.
• A clip built from references alone. Any model with image or video refs, such as MiniMax H3, Vidu Q4, HappyHorse 1.1 and Kling O3 (reference), the one under From references only. Vidu Q4 keeps people and things from up to 15 image refs, and voices from up to 3 audio refs. HappyHorse 1.1 makes 3 to 15 seconds from words, a start frame, or up to 9 image refs, at 720p or 1080p.
• A short clip from words. PixVerse v6, PixVerse v6 Max and Grok Imagine 1.5 make as little as 1 second, Kling 3.0 Turbo and HappyHorse 1.1 3, and MiniMax H3 4.
• A lighter Seedance. Seedance 2.0 Fast and Seedance 2.0 Mini take the same frames and references as Seedance 2.0, at 480p or 720p. Compare their prices in the table below.
• A clip you already have. The models under Work on a video, and Gemini Omni 1.1, which also makes clips from words, a start and an end frame, or up to 7 image refs: see Extend, edit or upscale a clip you have.

The models that make a clip

"From" is the shortest clip each model makes at its lowest resolution. Longer, sharper, or with sound where sound costs extra, costs more.

Shapes differ too. HappyHorse 1.1 offers the most, nine, from 9:21 to 21:9 with 4:5 and 5:4 among them; the Seedance models and MiniMax H3 offer 21:9 too; Kling 3.0 makes 16:9, 9:16 and 1:1, and Kling 4.0 and Kling 4.0 Flash add 21:9; Gemini Omni 1.1 makes 16:9 and 9:16. Output lists only the shapes the model you've picked makes.

The models that work on a clip

• PixVerse v6 (extend) carries a clip on, from 3.5 credits a second.
• Kling O3 (edit) changes a clip the way you say, from 11.5 credits a second.
• HappyHorse (edit) changes a clip, keeping its sound or making new, from 28 credits a second.
• Kling 3.0 Motion Control moves a character the way a clip moves, from 11.5 credits a second.

Comparing prices

The menu prices every model at the bar's current settings, and says what they are at its foot, such as "Prices at 5 s · 720p · sound". Change the length or resolution and every row follows. A model on a launch price says when it ends, and what it costs after.

Clips start from 2 credits, and a five-second Seedance 2.0 clip at 720p is 85 credits. The price on Generate is the one you pay. Models lists every model the site offers.