How the Visuall Canvas Works, Step by Step

Visuall turns a written idea into a finished video through a visible chain of steps on one canvas: you describe the film, the canvas agent splits it into scenes, you approve keyframes before anything is animated, then each approved frame is filmed with a model such as Veo 3.1, Seedance 2.0, or Gemini Omni. The practical payoff is that every step is a box you can see and change, and only the boxes you change get re-run — the rest is reused for free. This walkthrough covers the full path from an empty canvas to a stitched final video, plus where creators typically get stuck.

What the canvas actually is

Instead of a single prompt-to-video box, Visuall presents your project as a set of steps laid out on a canvas: your story, the AI scene plan, the keyframes, the model chosen for each shot, and the final stitch. The site describes it as "every video is a set of steps you can see and change."

That structure matters because it separates two things that are usually fused together in AI video tools: deciding what the shot looks like, and animating it. On the Visuall canvas those are different boxes, which is why you can approve a frame before spending anything on motion.

Step 1: Start from a template or describe the film

You have two entry points:

• Start from a template — the site shows examples like a product film, a vertical reel, a first-person powder run, and a short film set in Rome in 44 BC.
• Describe the film to the canvas agent — you write the story in plain language and the agent proposes the structure.

Either way, the output of this step is not video. It is a story box that the canvas will break down next.

Step 2: The story becomes a scene plan

The canvas agent splits your description into scenes with durations and shot descriptions. The site's skier example shows what this looks like:

Total: 10 seconds across 3 scenes. This is the box where you fix pacing problems cheaply — cutting a scene or rewriting a shot line here costs nothing compared to re-animating a clip.

Step 3: Generate and approve keyframes

Each scene gets a keyframe — a still image — before it becomes video. In the example, keyframes are generated with FLUX.2 at 16:9.

This is the approval gate. You look at the frame and decide: does this shot look right? If not, you change the prompt or the image and regenerate the keyframe. Only once you approve it does the frame move forward to animation.

For character consistency across scenes, the site's Rome example shows a related pattern: describe the character once in a paragraph, let the image model draw that face, then use an edit step to place the same face into each scene's frame. That keeps one character recognizable across four different locations.

Step 4: Animate the approved frames

Approved keyframes get filmed by a video model. The example uses Kling 3 at 5 seconds per shot, and the site names Veo 3.1, Seedance 2.0, and Gemini Omni as available models. Model choice is per shot, not per project — so you can use a cheaper or faster model for a simple wide shot and a stronger one for a hero shot.

If a scene needs dialogue or sound, Gemini Omni Flash is described as turning a frame into an 8-second clip that talks, with sound included.

Step 5: Re-run only what changed

This is the core mechanic and the reason the canvas is structured as boxes. When you change one shot, the other shots are not regenerated — they are reused. The site states it directly: "Change one shot, keep the rest," and "re-run only what changed; the rest is reused for free."

Practically, this means iterating on scene 2 does not cost you scenes 1 and 3 again. It also means the order of your work matters: lock the story and scene plan first, because changing the story re-runs the downstream boxes that depend on it.

Step 6: Stitch the final video

The last box joins the clips into one continuous video. In the Rome example, four clips become a single 32-second episode with the character's voice included. The whole project is summarized as 11 boxes, 225 credits, and about two minutes of waiting — a useful reference point for what a short narrative piece costs in steps and credits.

A worked example: the Rome episode

The site's behind-the-scenes breakdown is the clearest illustration of the full flow:

1. Describe her once — one paragraph about Mia; the image model draws the face she keeps in every shot.
2. One line per scene — four places in Rome, one per line; an edit step puts that same face into each frame.
3. Write what she says — each frame becomes an 8-second clip that talks, with sound.
4. Stitch the episode — the clips join into one 32-second video with her voice.

Notice that the character description and the scene list are separate inputs. That separation is what lets the same face appear in four different shots without re-describing her each time.

Where people get stuck

• Approving keyframes too fast. The approval gate is the cheapest place to fix a shot. A frame you dislike will not improve once animated.
• Editing the story after animating. Because downstream boxes depend on the story, late story edits re-run more than you expect. Lock the plan first.
• Treating model choice as global. Model selection is per shot on this canvas, so there is no reason to use one model for everything.
• Forgetting the reuse rule when budgeting. Re-runs are scoped to what changed, so cost scales with the number of boxes you touch, not the number of boxes in the project.

Getting started

The site offers a free start with 35 credits and states no credit card is required for that entry point. Current per-step costs are listed on the pricing page under "See what each step costs," which is the right place to check before committing to a multi-scene project, since credit use depends on how many scenes, keyframes, and animation runs you generate.