Video models
The video node lets you choose a video generation model for what you are making. Models differ in inputs, length, aspect ratio, resolution, and audio, so compare them before you choose.
What are the differences?
Video models differ in more than output quality — they differ in how the video is made. You can generate from a prompt alone, control the scene with start and end frames, or feed in a reference image or an existing video.
I want prompt-driven video in a range of aspect ratios.
Generating video while setting length and aspect ratio yourself across a wide range
Seedance 2.5Seedance 2.0 SeriesKling 4.0Coming soonMiniMax H3HappyHorse 1.1Coming soonI want to set the first and last scene myself.
Using a start frame and an end frame to control where the scene goes
Seedance 2.5Seedance 2.0 SeriesKling 2.6 / 3.0Kling OmniKling 4.0Coming soonMiniMax H3Veo 3.1 SeriesI want to set the scenes in between as well.
Setting the key scenes between the start and the end as keyframes to design how the story flows
Kling 4.0Coming soonI want to use reference images and an existing video together.
Connecting reference images and a video input at the same time to give the model more material to work from
Seedance 2.5Seedance 2.0 SeriesKling OmniKling 4.0Coming soonMiniMax H3I want the audio generated from the prompt along with the video.
Video audio generated automatically from the prompt instead of a separate sound on/off switch
Veo 3.1 SeriesGemini Omni FlashMiniMax H3HappyHorse 1.1Coming soonI want to make drafts quickly and generate them again and again.
Trying several drafts in a short time to settle on a direction
Seedance 2.0 MiniSeedance 2.0 FastKling 4.0 FlashComing soonVeo 3.1 FastModels at a glance
Compare the main input methods and generation options supported by the Klipse video node.
| Model | Specialty | Start | End | Reference | Video input | Length | Ratio | Resolution | Audio |
|---|---|---|---|---|---|---|---|---|---|
| Seedance 2.0 Mini | Fast, lightweight generation | Supported | Supported | Up to 9 | Supported | 4–15s | 6 ratios | 480p, 720p | Supported |
| Seedance 2.0 Fast | Fast repeated generation | Supported | Supported | Up to 9 | Supported | 4–15s | 6 ratios | 480p, 720p | Supported |
| Seedance 2.0 | High-resolution, general use | Supported | Supported | Up to 9 | Supported | 4–15s | 6 ratios | 480p, 720p, 1080p, 4K | Supported |
| Seedance 2.5 | Long clips, many references | Supported | Supported | Up to 30 | Supported | 4–30s | 6 ratios | 480p, 720p, 1080p | Supported |
| Kling 2.6 | Start and end frame control | Supported | Supported | Not supported | Not supported | 3–15s | Follows start frame | 720p, 1080p | Supported |
| Kling 3.0 | Frame control, high resolution | Supported | Supported | Not supported | Not supported | 3–15s | Follows start frame | 720p, 1080p, 4K | Supported |
| Kling Omni | Multi-input generation | Supported | Supported | Up to 7 | Supported | 3–10s | 3 ratios | 720p, 1080p, 4K | Supported |
| Kling 4.0 Coming soon | Long clips, multiple keyframes | Supported | Supported | Up to 10 | Supported | 3–30s | 4 ratios | 720p, 1080p, 4K | Supported |
| MiniMax H3 | 2K output, video references | Supported | Supported | Up to 9 | Supported | 4–15s | 6 ratios | 768p, 2K | Automatic, prompt-based |
| HappyHorse 1.1 Coming soon | Wide ratio choice, auto audio | Supported | Not supported | Up to 9 | Not supported | 3–15s | 9 ratios | 480p, 720p, 1080p | Automatic, prompt-based |
| Gemini Omni Flash | AI Edit, auto audio | Supported | Not supported | Up to 6 | Supported | 3–10s | 2 ratios | 360p, 720p, 1080p, 4K | Automatic, prompt-based |
| Veo 3.1 Pro | Frames, references, auto audio | Supported | Supported | Up to 3 | Not supported | 4, 6, 8s | 2 ratios | 720p, 1080p | Automatic, prompt-based |
| Veo 3.1 Fast | Fast frames and references | Supported | Supported | Up to 3 | Not supported | 4, 6, 8s | 2 ratios | 720p, 1080p | Automatic, prompt-based |
| Veo 3.1 Lite | Simple frame generation | Supported | Supported | Not supported | Not supported | 4, 6, 8s | 2 ratios | 720p, 1080p | Automatic, prompt-based |
Seedance 2.5
Generates the way the 2.0 series does, but a single clip can run twice as long and it accepts far more reference images.
Generates up to 30 seconds in one pass and accepts up to 30 reference images. It suits longer clips or work that needs many references. Resolution goes up to 1080p, and 4K is not supported. When a start frame is connected, the ratio follows the start frame.
- 4–30s
- Up to 30
- 9:16
- 16:9
- 1:1
- 21:9
- 4:3
- 3:4
- 1080p
Seedance 2.0 Series
Suits work that generates from a prompt alone, from start and end frames, or from reference images, while setting length and ratio freely.
Seedance 2.0 Mini
A lightweight model that supports 480p and 720p, suited to quick tests and low-resolution drafts.
Seedance 2.0 Fast
Supports 480p and 720p, suited to repeated generation and a fast production flow.
Seedance 2.0
Supports 480p through 4K, so it suits cases where the same generation method needs a higher output resolution.
Common to all: 4–15 sec · 9:16 / 16:9 / 1:1 / 21:9 / 4:3 / 3:4
Kling 2.6 / Kling 3.0
A setup built for specifying the first and last scene of a video directly as images. Neither model has a separate aspect ratio setting — both follow the ratio of the start frame image.
Kling 2.6
Supports 3–15 sec lengths and 720p or 1080p. It suits designing how the scene changes using a start frame and an end frame.
- Start Frame
- End Frame
- 3–15 sec
Kling 3.0
Supports 3–15 sec lengths and up to 4K. It suits wanting both start and end scene control and a high output resolution.
- Start Frame
- End Frame
- 3–15 sec
- 4K
Kling Omni
Along with start and end frames, it can take reference images and existing video as input, so it is built for video generation that draws on several pieces of media.
Video generation from multiple inputs
Because it can use reference images and video input together, it suits cases where you need to give the model more production information than a single image can carry. You can connect up to 7 reference images. With a video input connected, the limit is 4, and start and end frames are not available.
- Start Frame
- End Frame
- Reference Image
- Video Input
- 3–10 sec
- 9:16 / 16:9 / 1:1
- Up to 4K
Kling 4.0
Coming soonSuits making video of up to 30 seconds from a prompt alone or with start and end frames and a range of references added, and designing the scenes in between and the flow of the story with multiple keyframes.
Multiple keyframes + multiple references + up to 30 sec
You can set up to 10 keyframes to decide a character's state, scene changes, and the key moments of the story. A keyframe is an input that sets a key scene of the video, which makes it different from a reference image the video draws a character or style from.
References can be images, videos, elements such as a person or product, and a voice. You can use up to 10 reference images, up to 5 videos (30 seconds in total), and up to 7 elements (up to 3 of them video-based), with up to 15 in all. Video-based elements count toward the total video length, and a voice is used as a reference for how the speech sounds.
Supports lengths of 3–30 seconds, 720p, 1080p, and 4K resolution, and the 9:16, 16:9, 1:1, and 21:9 ratios. Audio comes out as two-channel stereo, and it can generate speech and on-screen text in several languages, including Korean.
It also supports editing an existing video — changing expressions, motion, camera, background, style, and more. You can input up to 5 clips, including the one video to edit; each clip can be 3–30 seconds, up to 30 seconds in total.
- Start Frame
- End Frame
- Up to 10 keyframes
- Reference Image
- Video Input
- Voice Reference
- 3–30 sec
- 4 ratios
- Up to 4K
- Stereo Audio
Kling 4.0 Flash
Coming soonA Kling 4.0 family model suited to making drafts quickly and generating them again and again.
Fast repeat generation + up to 20 sec
Makes video from text or an image, and can also generate with references. It generates 3–20 second videos at 720p.
- Image to Video
- Reference
- 3–20 sec
- 720p
MiniMax H3
Generates stereo audio automatically, and suits making video from a prompt alone or with start and end frames or reference images and video inputs added.
Multiple references + stereo audio
An omni-modal model that reads text, images, and video in one pass. A character can come from a reference image, and the scene and movement from a video input.
You can choose a length of 4–15 seconds and 768p or 2K resolution, and the supported ratios are 9:16, 16:9, 1:1, 21:9, 4:3, and 3:4. You can connect up to 9 reference images and up to 3 video inputs as references, and use both at once. With reference images or video inputs connected, start and end frames are not available. When a start frame is connected, the ratio follows the start frame. Audio has no on/off setting and is generated automatically from the prompt, as native stereo created together with the video.
- Start Frame
- End Frame
- Reference Image
- Video Input
- 4–15 sec
- 6 ratios
- 768p / 2K
- Audio Auto
HappyHorse 1.1
Coming soonSupports nine ratios and automatic audio, and suits making video from a prompt alone or with a start frame or reference images added.
Wide ratio choice + automatic audio
Generates video from a prompt alone, or from a start frame or reference images, keeping a character, product, outfit, or brand element in the same shape and style through the video. It suits work where the same person or product has to appear again across several videos.
You can choose a length of 3–15 seconds and 480p, 720p, or 1080p resolution, and the supported ratios are 9:16, 16:9, 1:1, 21:9, 9:21, 4:3, 3:4, 4:5, and 5:4. You can connect one start frame or up to 9 reference images, but not both at once. End frames and video input are not supported. When a start frame is connected, the ratio follows the start frame. Audio has no on/off setting and is generated automatically from the prompt, with lip sync matched automatically.
- Start Frame
- Reference Image
- 3–15 sec
- 9 ratios
- Up to 1080p
- Audio Auto
Gemini Omni Flash
Suited to creating video from a start frame and reference images or using AI Edit on a video input, with audio generated automatically from the prompt.
Multiple inputs + automatic audio
You can choose a length of 3–10 seconds, the 9:16 or 16:9 ratio, and 360p, 720p, 1080p, or 4K resolution. Video input is used for AI Edit only, and the result then follows the input video's length and ratio. Audio has no on/off setting and is generated automatically from the prompt.
- Start Frame
- Reference Image
- Video Input
- 3–10 sec
- 9:16 / 16:9
- Up to 4K
- Audio Auto
Veo 3.1 Series
Suited to specifying start and end frames and generating the audio from the prompt at the same time. Pro and Fast can use reference images instead of a start frame, while Lite offers a simpler input structure.
Veo 3.1 Pro
Suits making video from start and end frames or from reference images. Start and end frames cannot be combined with reference images, and using reference images fixes the length at 8 seconds.
Veo 3.1 Fast
A good choice when you want the same input structure as Pro but are working toward fast repeated production.
Veo 3.1 Lite
Suits making video from start and end frames and the prompt, without a reference image.
Common to all: 4 / 6 / 8 sec · 9:16 / 16:9 · 720p / 1080p · audio generated automatically from the prompt
Choosing a model in the video node
The actual video generation happens in the video node in Canvas. When you select a model, the input slots, length, ratio, resolution, and sound options it supports change automatically.
What the video node and AI Models each cover
The video node page explains how to connect inputs and generate video in practice, while AI Models · video models compares what each model specializes in and how to choose.
Frequently asked questions
Which model is best?
No single model is best for every task. Choose based on the input method, video length, aspect ratio, resolution, and audio generation you need.
Which models give the widest choice of aspect ratio?
Seedance 2.5, the Seedance 2.0 Series, and MiniMax H3 support 6 ratios (9:16, 16:9, 1:1, 21:9, 4:3, 3:4), and HappyHorse 1.1 supports 9 (9:16, 16:9, 1:1, 21:9, 9:21, 4:3, 3:4, 4:5, 5:4). Kling 4.0 supports 4 (9:16, 16:9, 1:1, 21:9), Kling Omni supports 3 (9:16, 16:9, 1:1), and Gemini Omni Flash and the Veo 3.1 Series support 2 (9:16, 16:9). Kling 2.6 and Kling 3.0 have no separate ratio setting and follow the ratio of the start frame image.
Which model should I choose to use both a start frame and an end frame?
Seedance 2.5, the Seedance 2.0 Series, Kling 2.6, Kling 3.0, Kling Omni, Kling 4.0, MiniMax H3, and the Veo 3.1 Series support both a start frame and an end frame. Kling 4.0 also lets you set up to 10 keyframes to decide the scenes in between. HappyHorse 1.1 and Gemini Omni Flash support a start frame only.
Can I use a reference image and an existing video together?
Seedance 2.5, the Seedance 2.0 Series, Kling Omni, Kling 4.0, and MiniMax H3 let you connect reference images and a video input at the same time. Gemini Omni Flash cannot use reference images while a video input is connected, and Kling 2.6, Kling 3.0, HappyHorse 1.1, and the Veo 3.1 Series do not support video input.
Can the sound be generated automatically along with the video?
The Veo 3.1 Series, Gemini Omni Flash, MiniMax H3, and HappyHorse 1.1 generate audio automatically from the prompt, with no sound on/off setting. Seedance 2.5, the Seedance 2.0 Series, Kling 2.6, Kling 3.0, and Kling Omni have a sound on/off setting you choose yourself. Kling Omni does not generate new audio while a video input is connected. As a reference the sound stays off, and for AI Edit you choose whether to keep the original video's sound.
Which models support 4K?
Seedance 2.0, Kling 3.0, Kling Omni, Kling 4.0, and Gemini Omni Flash support resolutions up to 4K. MiniMax H3 goes up to 2K, Seedance 2.5, Kling 2.6, HappyHorse 1.1, and the Veo 3.1 Series up to 1080p, and Seedance 2.0 Mini, Seedance 2.0 Fast, and Kling 4.0 Flash up to 720p.
I need a long clip, up to 30 seconds.
Seedance 2.5 generates 4–30 seconds and Kling 4.0 generates 3–30 seconds in a single pass. Kling 4.0 Flash generates 3–20 seconds, the Seedance 2.0 Series and MiniMax H3 generate 4–15 seconds, Kling 2.6, Kling 3.0, and HappyHorse 1.1 generate 3–15 seconds, Kling Omni and Gemini Omni Flash generate 3–10 seconds, and the Veo 3.1 Series offers 4, 6, and 8 seconds.
How do Kling 4.0 and Kling 4.0 Flash differ?
Kling 4.0 makes video of up to 30 seconds and 4K, and suits designing scenes in detail with multiple keyframes and a range of references. Kling 4.0 Flash makes 3–20 second videos at 720p quickly, and suits generating drafts again and again.