Video Generation

The Video Generation node generates a video from a text prompt, a start frame image, or reference footage, depending on the model you pick.

Inputs and outputs

Most models include Start Frame (the opening frame) and Prompt. The exception is Kling 3.0 Motion Standard, which uses a character image and a motion video instead (see below). Depending on the model, you may also see:

End Frame: the closing frame, for models that support it
Image Reference: images showing the look you want the model to follow
Elements: your Influencer's images, so the same character shows up in the clip
Reference Video: a clip the model takes its cues from, for models that support it

Image Reference and Elements both feed the model reference images. They are separate inputs because they do different jobs: one sets the look, the other sets who is in the shot.

Each input only connects to what it's named for: images go on Image Reference and Elements, a video goes on Reference Video. A Sequence can't connect to this node at all.

Plus a generic Flow input.

Output: a single Video handle.

Writing a prompt

Type your prompt into the box at the bottom of the card ("Describe the video scene..."). If Prompt is connected from elsewhere, this box is disabled and shows "Prompt connected from input" instead.

You can connect more than one LLM, Text or Agent node to Prompt. For example, one writes the scene and another writes the movement. Their texts are joined into one prompt, in the order you connected them.

The generated video plays in the preview area at the top with standard playback controls. Click Download in the settings bar to save it.

Click Save to Media, next to Download, to keep a copy in your Media Library, in the folder named after your workflow. It saves the video the card is showing, and the icon turns into a check once it's saved. If your storage is full, the card tells you and nothing is saved.

Choosing a model

Click the model picker to search and choose a model. The default model is Seedance 2.0 Fast, with a default duration of 5 seconds, aspect ratio 9:16, and sound generation off by default.

The models run on Atlas Cloud, like every node.

If you connect Image Reference, Elements, Reference Video, or Reference Audio, the model picker filters itself down to models that can use that input.

Most models make a video either from your frames or from your references, never from both at once. On the Seedance models, MiniMax H3, Gemini Omni Flash 1.1 and Gemini Omni Flash, Wan 3.0, HappyHorse 1.1 and 1.0, and PixVerse V6, Start Frame and End Frame can't be used together with Image Reference, Reference Video or Reference Audio. Once you connect one of them, the others turn grey, and hovering one tells you what to remove. A workflow that already has both connected won't start until you disconnect one, so nothing is spent on a video that ignores half of what you gave it.

Start Frame and End Frame each take one picture, and an End Frame is only used together with a Start Frame. Once one is connected it won't take a second, and hovering it tells you to disconnect it first. A workflow that already has two pictures on one frame won't start until you disconnect one. Image Reference is different: you can connect several nodes to it, and the video uses all of their pictures, up to the number the model accepts.

Settings on the card

Shown only when the selected model supports them: Model picker, Aspect ratio chip, Duration chip (for example "5s" or "8s"), Quality chip, and a Sound on/off toggle.

Quality and aspect ratio work together. Pick the quality first, because it decides which shapes are available: a model may offer an ultra-wide 21:9 at 720p but not at 4K. The two choices resolve to an exact frame size behind the scenes, so you never have to think in pixels.

Models available

This node runs on Atlas Cloud. You can choose from MiniMax H3 Max and H3, Gemini Omni Flash 1.1 and Gemini Omni Flash, Wan 3.0, Seedance 2.5, 2.0, 2.0 Mini and 2.0 Fast, HappyHorse 1.1 and 1.0, Kling 3.0 Standard, Kling 3.0 4K, Kling O3 4K, Kling O1, Kling 3.0 Motion Standard, Cosmos 3 and PixVerse V6. Each model supports its own mix of text-to-video, image-to-video, or reference-to-video, with its own durations, quality tiers, and aspect ratios; check the model picker for what each one offers. Clip lengths run from 1 second to 30 depending on the model.

Cosmos 3 only animates a photo, so connect a start frame. MiniMax H3 Max, Cosmos 3, Kling 3.0 Standard, Kling 3.0 4K, Kling O3 4K and Kling O1 have no reference mode; the first three Kling models take Elements instead. Kling O1 has no sound and makes clips of 5 or 10 seconds. The Seedance models, Wan 3.0 and MiniMax H3 also take reference video and audio, and Gemini Omni Flash 1.1 takes reference video. Quality options marked "sr" or "esr" are upscaled from a smaller native video and cost more. The price next to each model in the picker is what one clip costs at the node's starting settings, for example "$0.12 / 5 s clip". Other lengths, qualities and modes cost more or less; the Model prices page lists them all. Kling 3.0 Motion Standard shows a price per second, because its length follows your motion video.

Kling 3.0 Motion Standard works a bit differently: you provide one character image and one motion video, and the model transfers the motion video's movement onto your character. There's no duration or aspect ratio to set, the length follows the motion video.

Running the node

Select the node to reveal its Run button. A standalone node with no inputs runs itself; run it again to also run everything downstream. If its inputs already have results, running it runs everything downstream too. If inputs aren't ready yet, you'll be asked whether to run the whole workflow, or run everything up to and including this node.

Click the Inputs / Outputs button at any time to see exactly what data went in and came out.