Video

The Video tab makes new clips. Choose how you guide the clip under Input Mode in the settings panel. Each input mode has its own inputs, and the model picker only offers the models that support it.

To change a clip you already have (edit it, extend it or change its ratio), use Video Edit.

Make a clip

1. Pick an input mode

   Video opens on Start Frame with MiniMax H3 Max. Choose Text to Video to work from the prompt alone, or another input mode from the list.

2. Choose a model

   Click the model card at the top of the settings panel to open the picker. Changing the input mode can change the model, because not every model supports every mode. The models table shows which input modes each model covers.

3. Load inputs and write the prompt

   Inputs go in the panel on the left. To start from an image in the Gallery, right-click it and send it to Video Gen: it arrives as the starting frame. Then describe the action, the motion and the camera in the prompt bar. Type @ to insert a snippet.

4. Set the output and generate

   The ratio, resolution and duration pills only offer what the model supports. Some inputs set the length for you, such as an audio track or a motion clip. Click Generate: the button shows the credit cost first.

Start Frame and Start + End Frame

Start Frame animates one image: load it in Starting Frame and describe what happens next. Start + End Frame adds an Ending Frame, and the clip moves from the first image to the second.

• A few models, such as Kling 3.0 Omni, also take reference images next to the frames.
• Kling and Seedance only accept frames within a range of ratios. If yours is outside it, the input card asks you to crop it.

References

References drives the clip with media you load as Reference 1, Reference 2 and so on: images, and on some models video or audio. Name them in the prompt with @ref1, @ref2. Dropping several files on one slot fills the next ones.

• Limits vary by model. Kling 3.0 Omni takes up to seven images, or four images and one video. MiniMax H3 takes up to 9 images, 3 videos and 3 audio clips, 12 files in all.
• Seedance 2.0 and later refuse references that show a real person’s face. A face generated with Seedream 5.0 or later is accepted, so you can build the character there first.

Avatar

Avatar animates a character image with a voice or a song. Load the image in Starting Frame and the performance in Audio. The clip’s length follows the audio. Two models offer it:

• Kling Avatar works best with one clear character image and one spoken audio file: MP3, WAV, M4A or AAC, from 2 to 300 seconds.
• OmniHuman 1.5 takes from 1 to 60 seconds of audio.

You can make the audio in Voice or Music.

Keyframes (Beta)

Keyframes pins images to points in time and renders a clip that passes through them in order. Each slot (Keyframe 1, Keyframe 2 and so on) has a Timecode control. Add slots as you need them.

• Ray 3.2 takes up to 64 keyframes, Flux 3.0 Video up to 10. Keyframes opens on Flux 3.0 Video.
• Keyframes is only available in Create. Nodes does not offer it.

Motion Control (Beta)

Motion Control copies the movement of a video clip onto your image. Load the image in Starting Frame and the clip in Motion Reference. It runs on Kling 3.0.

• The motion clip must be 3 to 10 seconds long, and it sets the length of the result. The ratio follows the starting frame.
• It works best with one clearly visible person making clear movements, and a starting frame that matches the first pose of the clip. Clips with no person, little movement, cuts or heavy camera motion may fail.

Sound

Many models can generate sound with the clip. When the selected model can, a Generate Audio switch appears in the settings panel. On a few models the sound is always on.

Some models also take audio as an input. LTX can drive a Start Frame clip from an Audio file, and Wan 3.0, Seedance and MiniMax H3 accept audio among their references.

Multi-shot prompts

Several models can render one clip as a sequence of shots, among them Kling 3.0, Seedance, MiniMax H3 and Wan 3.0. With one of them selected, click Builder in the prompt bar: the Prompt Builder shows a shot timeline, locked to the duration you picked.

Give each shot its own description and length, and link a reference or a camera snippet to it. The builder writes the prompt in the format the active model expects, so you can render the same shots on another model without rewriting them.

Draft and Finalize

Flux 3.0 Video, MiniMax H3 and Seedance 2.5 can render a Draft: a cheaper preview, at 720p on Flux, 768p on H3 and 480p on Seedance 2.5. Turn on Draft in the settings panel, and the Generate button changes to Draft.

When a draft is worth keeping, open it and click Finalize to render it again at full quality: 1080p on Flux and Seedance 2.5, 1440p on H3. The result is usually close to the draft, but fine detail or motion can shift. H3 and Seedance 2.5 drafts can be finalized for up to a week after they were made. See Pricing for costs.

Model settings

Some models add their own settings under the model card, most of them inside Advanced Controls. For example:

Models

Every model in the Video tab, with the input modes, resolutions and durations it supports. The longest clips, up to 30 seconds, come from Wan 3.0 and Seedance 2.5.

MiniMax

• H3

  Multimodal video up to 1440p from text, frames, mixed references, or prompt-based video edits; drafts at 768p

  Text to Video, Start Frame, Start + End Frame, References768p, 1440p4 to 15s

• H3 MaxDefault

  H3 variant tuned for stronger prompt adherence; text, frame, or reference-driven video at 480p/768p, 5-15s

  Text to Video, Start Frame, Start + End Frame, References480p, 768p, 1080p5 to 15s

• H3 Max Turbo

  H3 Max on a faster inference stack: same render, half the price, no reference support

  Text to Video, Start Frame, Start + End Frame480p, 768p, 1080p5 to 15s

VeoGoogle

• Veo 3.1 Lite

  Most affordable Veo 3.1 option

  Text to Video, Start Frame, Start + End Frame720p, 1080p4, 6, 8s

• Veo 3.1 Fast

  Quick previews with good quality

  Text to Video, Start Frame, Start + End Frame, References720p, 1080p, 4K4, 6, 8s

• Veo 3.1

  Full quality generation

  Text to Video, Start Frame, Start + End Frame, References720p, 1080p, 4K4, 6, 8s

GeminiGoogle

• Omni 1.1 FlashPreview

  Fast generation, interpolation, editing, and continuation at 360p, 720p, 1080p, or 4K.

  Text to Video, Start Frame, Start + End Frame, References360p to 4K3 to 10s

SeedanceBytePlus

• Seedance 2.5

  Up to 30s in one shot, 50 references, timestamped editing

  Text to Video, Start Frame, Start + End Frame, References480p, 720p, 1080p4 to 30s

• Seedance 2.0

  Multimodal generation, edit, and extend

  Text to Video, Start Frame, Start + End Frame, References480p, 720p, 1080p4 to 15s

• Seedance 2.0 Fast

  Faster multimodal generation, edit, and extend model

  Text to Video, Start Frame, Start + End Frame, References480p, 720p4 to 15s

• Seedance 2.0 Mini

  Cheapest multimodal generation, edit, and extend model

  Text to Video, Start Frame, Start + End Frame, References480p, 720p4 to 15s

KlingKlingAI

• Kling 3.0 Turbo

  Fast text/image video with multi-shot, 720p/1080p

  Text to Video, Start Frame720p, 1080p3 to 15s

• Kling 3.0

  Core text/image video with multi-shot and motion control

  Text to Video, Start Frame, Start + End Frame, Motion Control720p, 1080p, 4K3 to 15s

• Kling 3.0 Omni

  Omni mixed-media video with references

  Text to Video, Start Frame, Start + End Frame, References720p, 1080p, 4K3 to 15s

• Kling Avatar

  Talking avatar from a start frame and audio

  Avatar720p, 1080p

Luma

• Ray 3.2

  HDR-capable video with keyframes, edit and expand

  Text to Video, Start Frame, Start + End Frame, Keyframes540p, 720p, 1080p5, 10s

GrokxAI

• Imagine Video

  Text, start frame, references, edit, and extend

  Text to Video, Start Frame, References480p, 720p1 to 15s

• Imagine Video 1.5

  Text, start and end frames with references, keyframes and preset voices; 1080p for text and start frames without references

  Text to Video, Start Frame, Start + End Frame, Keyframes, References480p, 720p, 1080p1 to 15s

• Imagine Video 1.5 Lite

  Text and start frame at a lower price, up to 1080p

  Text to Video, Start Frame480p, 720p, 1080p1 to 15s

FluxBlack Forest Labs

• Flux 3.0 Video

  Draft-capable video with audio (720p/1080p, 5–20s)

  Text to Video, Start Frame, Keyframes720p, 1080p5 to 20s

HappyHorseAlibaba

• HappyHorse 1.1

  Physically realistic text-to-video

  Text to Video, Start Frame, References720p, 1080p3 to 15s

LTX

• LTX 2.5 Fast

  Native multi-shot, up to 4K and 20s

  Text to Video, Start Frame, Start + End Frame720p to 4K6 to 20s

• LTX 2.5 Pro

  Native multi-shot at the highest fidelity, up to 4K

  Text to Video, Start Frame, Start + End Frame720p to 4K6, 8, 10s

• LTX 2.3 Fast

  Quick iteration, longer durations up to 20s

  Text to Video, Start Frame, Start + End Frame1080p, 1440p, 4K6 to 20s

• LTX 2.3 Pro

  Higher fidelity, audio-driven & interpolation

  Text to Video, Start Frame, Start + End Frame1080p, 1440p, 4K6, 8, 10s

OmniHumanBytePlus

• OmniHuman 1.5

  Image + audio to talking portrait video

  Avatar720p, 1080p

WanAlibaba

• Wan 3.0Beta

  All-in-one: text, start/end frame, and mixed image/video/audio references up to 30s

  Text to Video, Start Frame, Start + End Frame, References480p, 720p, 1080p2 to 30s

• Wan 3.0 PrimeBeta

  Same capabilities as Wan 3.0 with significantly faster generation, at a higher rate

  Text to Video, Start Frame, Start + End Frame, References480p, 720p, 1080p2 to 30s