Free Canvas: Video Generation & Reference Video

1. Preparation

1. Models: Settings → Models → add and enable a provider (Volcengine Ark, Kling, MiniMax, OpenRouter, …) → tick the models you want on the Video tab and set the default video model.
2. Object storage (required for reference video): Settings → Object Storage → add and enable one of Volcengine TOS / Alibaba Cloud OSS / Tencent Cloud COS. A local reference video is first uploaded to a publicly reachable URL (a signed URL for private buckets) so the model can fetch it.
3. Make sure the provider account has credit — submission fails immediately when the balance is low (e.g. MiniMax 1008).

2. Create a free canvas

1. With the project open, create a new Free Canvas from the workbench toolbar or the asset library.
2. If a “Free Canvas” already exists, new ones are named “Free Canvas 2”, “Free Canvas 3”… The name only distinguishes them in the asset library; you can rename the asset there.
3. Double-click the Free Canvas asset to enter the blank node graph (host inner graph).

3. Build a minimal video chain

Right-click on empty space in the Free Canvas graph (or press Space for the menu) and add / connect nodes in this order:

1. Add a Video Generation node. Pick the video model on the node or in the Inspector, then write the generation prompt.
2. (Optional) Add a Text / Note node, or an upstream Prompt Optimizer, and connect its text output to the Video Generation Text input.
3. (Optional) Add Select Video, or drag a video asset from the asset library as a host / reference node, and connect its video output to the Video Generation Video input (reference video).
4. (Optional) Connect an image node or asset to the Image / First frame / Last frame port (depends on the model and frame mode — see the next section).
5. Add a Video Output node and wire the generation output into it for preview and export.

Sketch (text-to-video + reference video):

[Text / Prompt] ─────► [Video Generation] ─────► [Video Output]
                                        ▲
[Select Video / Video Asset] ──────────┘   (Video port = reference video)

4. How to connect a reference video

The Video Generation node's Video input is for a reference video: the model can borrow motion, camera work, character actions or scene pacing (actual capability varies by provider and model).

• Source: videos imported into the asset library, a Select Video node, upstream generated videos, or another host's video output.
• Upload: at run time, if the reference is still a local path / data URL, the app uploads it through the currently enabled object storage and passes the URL to the model API.
• Cleanup: temporary uploaded references are usually deleted once the task finishes, so test files don't pile up in the bucket.
• Multiple references: where the Video port accepts several links, you can attach multiple reference clips (subject to actual model support).

5. First/last frames and prompts

• Pick the frame mode in the Video Generation node params:
  • None: no forced first/last frame (pure text-to-video, or reference video / image only).
  • First frame: exposes a first-frame image port — good for image-to-video and frozen openings.
  • First + last frame: constrains both the opening and closing shot (requires model support, e.g. some Hailuo / Kling capabilities).
• Use @ in the prompt to reference connected upstream media (same as image generation) and describe motion, camera and pacing.
• Duration, resolution and audio options depend on the current model capability catalog; unsupported options are ignored or disabled.
• Frame Pulling: connect an upstream video to the Frame Pulling node and double-click to open the pull station — browse frame by frame on the filmstrip, grab key frames and annotate them; the grabbed frames can be wired into downstream generation nodes as references.
• Segment Reshoot: connect the source video to the Segment Reshoot node and double-click to open the reshoot station — mark the start / end time to change and describe the change (e.g. “make the umbrella transparent”); running it reshoots only that range (Seedance 2.5 timestamp-level editing) and leaves the rest untouched. The prompt box supports @ references to port media, and you can pick the video model on the right.

6. Run and inspect results

1. Select the Video Generation or Video Output node → click Execute in the toolbar, or hold C to open the execution ring.
2. You can also Add to Tasks on the output node and watch background progress in the task list; upstream nodes shared by several chains are cooked only once.
3. On success, play it back in the node preview area: Back to start on the left, the progress bar in the middle, and Play / Pause on the right.
4. To assemble a final cut, wire the result into the script asset's Final Timeline, or export the media file.

7. FAQ

For the full interface reference see the Manual; model and object storage details are in chapter 4 of the manual.