AI Video Model Spec Sheet
17 AI video models side by side: how long a clip can be, which resolutions they sell and at what price per second, whether they make sound, speak dialogue, take an end frame or reference images.
Specifications
| Model | Clip length | Resolution and price per second | Sound | Speaks dialogue | End frame | Reference images | Aspect ratios | Good to know |
|---|---|---|---|---|---|---|---|---|
| Kling 3.0 ProKuaishou | 5, 10 s | 1080p $0.11 | OptionalPriced with sound off | No | No | No | 16:9, 9:16, 1:1 | Sold in 5 s and 10 s clips only, so a 6 s shot is billed as 10 s. Price shown is with sound off. |
| Kling 3.0 StandardKuaishou | 5, 10 s | 720p $0.084 | OptionalPriced with sound off | No | No | No | 16:9, 9:16, 1:1 | Sold in 5 s and 10 s clips only. Price shown is with sound off. |
| Kling 2.5 Turbo ProKuaishou | 5, 10 s | 1080p $0.070 | No | No | No | No | 16:9, 9:16, 1:1 | Sold in 5 s and 10 s clips only. |
| Seedance 2.5ByteDance | 4 to 30 s | 480p $0.22 720p $0.47 |
Optional | Yes | Yes | Up to 50 | Follows the image | Up to 30 s in one take, the longest single clip here. Turns away photoreal human faces; best for places, products, animals and stylised people. Tops out at 720p. |
| Seedance 2.0ByteDance | 4 to 15 s | 720p $0.30 1080p $0.68 |
Optional | No | Yes | Up to 9 | 16:9, 9:16, 4:3, 3:4, 1:1, 21:9 | Turns away photoreal human faces. |
| Seedance 2.0 FastByteDance | 4 to 15 s | 720p $0.24 | Not published | No | Yes | Up to 9 | 16:9, 9:16, 4:3, 3:4, 1:1, 21:9 | Turns away photoreal human faces. |
| MiniMax H3MiniMax | 5 to 15 s | 480p $0.050 768p $0.060 2K $0.13 4K $0.16 |
Always on | Yes | Yes | Up to 9 | Follows the image | Native 2K with synced sound. Reference mode: first 5 images free, then $0.08 each. |
| LTX 2.3 ProLightricks | 6, 8, 10 s | 1080p $0.080 | Optional | No | Yes | No | 16:9, 9:16 | 1080p is the lowest tier it renders. Sold in 6, 8 and 10 s clips. |
| LTX 2.3 FastLightricks | 6, 8, 10 s | 1080p $0.060 | Optional | No | Yes | No | 16:9, 9:16 | Sold in 6, 8 and 10 s clips. |
| LTX 2.5 ProLightricks | 6, 8, 10 s | 720p $0.12 1080p $0.17 |
Optional | Yes | Yes | No | 16:9, 9:16 | Sold in 6, 8 and 10 s clips. |
| LTX 2.5 FastLightricks | 6 to 20 s in 2 s steps | 720p $0.090 1080p $0.13 1440p $0.19 4K $0.30 |
Optional | Yes | Yes | No | 16:9, 9:16 | Fast means quicker to render, not a draft: it reaches 20 s and 4K. Sold in 2 s steps from 6 s. |
| Wan 2.7Alibaba | 2 to 15 s | 720p $0.10 1080p $0.15 |
No | No | No | No | 16:9, 9:16 | None |
| Wan 3.0Alibaba | 2 to 30 s | 480p $0.050 720p $0.10 1080p $0.20 |
Optional | Yes | Yes | Up to 10 | 16:9, 4:3, 1:1, 3:4, 9:16 | Up to 30 s in one take. Takes no negative prompt. |
| Wan 3.0 PrimeAlibaba | 2 to 30 s | 480p $0.068 720p $0.14 1080p $0.28 |
Optional | Yes | Yes | Up to 10 | 16:9, 4:3, 1:1, 3:4, 9:16 | About 40% dearer than Wan 3.0; the vendor publishes no quality claim for the difference. |
| FLUX 3Black Forest Labs | 5 to 20 s | 720p $0.17 1080p $0.29 |
Optional | Yes | Yes | No | 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16 | Lip-sync in 13 languages. Keyframe mode pins up to 10 stills on the timeline. |
| FLUX 3 DraftBlack Forest Labs | 5 to 20 s | 720p $0.060 | Not published | Yes | Yes | No | 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16 | A cheap 720p preview tier for testing motion before the full render. |
| Gemini Omni 1.1 FlashGoogle | 3 to 10 s | 360p $0.030 720p $0.10 1080p $0.15 4K $0.30 |
Always on | Yes | Yes | No | 16:9, 9:16 | The cheapest second here at 360p. 16:9 and 9:16 only. Sound is always on; no seed or negative prompt. |
Prices are API list prices in US dollars per second of output, checked against each provider's published price list on October 22, 2026. Vendors: Kuaishou, ByteDance, MiniMax, Lightricks, Alibaba, Black Forest Labs, Google. Data as JSON; free to reuse with a link back to this page.
Quick Answers
Worked out from the table above, so they change when the models do.
Which AI video models generate sound?
MiniMax H3 and Gemini Omni 1.1 Flash always render a sound track. Kling 3.0 Pro, Kling 3.0 Standard, Seedance 2.5, Seedance 2.0, LTX 2.3 Pro, LTX 2.3 Fast, LTX 2.5 Pro, LTX 2.5 Fast, Wan 3.0, Wan 3.0 Prime and FLUX 3 can add sound when you ask for it. Kling 2.5 Turbo Pro and Wan 2.7 are silent.
Which AI video models can speak dialogue with lip-sync?
Seedance 2.5, MiniMax H3, LTX 2.5 Pro, LTX 2.5 Fast, Wan 3.0, Wan 3.0 Prime, FLUX 3, FLUX 3 Draft and Gemini Omni 1.1 Flash. Each wants the spoken line written in its own format; the prompt translator writes it for you.
Which AI video model makes the longest single clip?
Seedance 2.5, Wan 3.0 and Wan 3.0 Prime, at up to 30 seconds in one generation. Longer shots on other models have to be made as several clips and joined.
What is the cheapest AI video model per second?
Gemini Omni 1.1 Flash at 360p, $0.030 per second. At 1080p the cheapest is LTX 2.3 Fast at $0.060 per second, and the middle of the field is about $0.15.
Which AI video models take an end frame?
Seedance 2.5, Seedance 2.0, Seedance 2.0 Fast, MiniMax H3, LTX 2.3 Pro, LTX 2.3 Fast, LTX 2.5 Pro, LTX 2.5 Fast, Wan 3.0, Wan 3.0 Prime, FLUX 3, FLUX 3 Draft and Gemini Omni 1.1 Flash. An end frame lets you decide where a shot finishes, which makes cuts between shots line up.
Which models take reference images for consistent characters?
Seedance 2.5 (up to 50), Wan 3.0 (up to 10), Wan 3.0 Prime (up to 10), Seedance 2.0 (up to 9), Seedance 2.0 Fast (up to 9), MiniMax H3 (up to 9), through their reference-to-video modes.
Which AI video models refuse real faces?
Seedance 2.5, Seedance 2.0 and Seedance 2.0 Fast turn away photoreal human faces. They are strong for places, products, animals and stylised characters.
Plan the whole production, not just the clips.
PyxelJam Studio takes a script to a finished video, with an estimate before every step and your approval at each one.
Write it or import it
Framing, moves, lengths
A drawn frame per shot, same cast every time
Approved frames become video
Trim, order, music, title card
Export or send to your channels
Free to start. No card needed.