6 Things You Can Make with Blender and Seedance 2.5

TL;DR: You can use rough 3D models in Blender to guide AI video generation. We look at six practical shots: live-action previs, gallery installation previews, miniature reveals, character action beats, multi-style tests, and game cutscenes. Making a simple gray blockout in Blender lets you test camera moves with Seedance 2.5 before you spend credits on final video clips.

Pre-visualization (previs) means sketching out camera motion, shot speed, and subject layout before final production. When you build a simple scene in Blender, you give the AI video model a visual map.

That map shows the model where the camera moves and where objects stand. It's much easier than guessing how a text description might turn out.

Our tests used Blender 5.2.1 LTS and Seedance 2.5 at 480p across five-second clips. We built three original five-second gray scenes at 24 fps (120 frames total). We ran eight total generations to see what works and what drifts.

We also reviewed two historical test clips to see how the system handles longer shots and fast character action.

What a Blender reference adds to your prompt

A text prompt tells the AI what materials, subjects, and lights should look like. But describing complex camera paths in words can leave room for surprise interpretations.

When you export a gray clay pass from Blender, you give the model a clear visual guide. Documented white-model guidance for video generation shows that simple gray shapes communicate where objects sit, camera paths, and subject placement before adding lighting and textures. You can read our prompt guide to see how this works in detail.

With the Blender add-on, you can set keyframes on your camera, export a local gray pass, and upload that video file.

The add-on doesn't send raw 3D mesh points or camera coordinates to the cloud. Instead, it uploads your rendered video file. The AI model watches that moving reference and paints new materials, lighting, and textures over the top.

Here's how the process moves from your viewport to the final output:

1. Build simple shapes: Place basic gray boxes and stand-in shapes in your Blender scene.
2. Keyframe the camera: Animate your camera move and check framing in your viewport.
3. Render a local clay pass: Export a short MP4 reference clip without spending credits.
4. Send to Seedance: Upload your video reference with a text prompt and look images.
5. Review the result: Download the finished clip or adjust your Blender keyframes to try again.

1. Previs for a live-action shot

Planning a live-action camera move can be tricky. A director often needs to plan drone paths, crane sweeps, and actor positions before renting expensive gear.

Building a rough 3D scene lets you test whether a sweeping camera move feels right. You'll see the speed of a low-altitude pass and know if the framing works.

To see how the model handles a long flight path, we reviewed a historical establishing drone flight (test b2b93be2).

Historical b2b93be2 clip: Seedance 2.0 Mini, 480p, 14.08-second output generated from a 13.75-second cropped public Blender screen recording and a reference image.

This historical test used a 13.75-second cropped Blender flyover as a reference video, paired with an appearance image. The prompt read:

Real medieval countryside drone footage. Follow @Video1 for camera motion; use @Image1 for visual appearance.

The 14.08-second output followed the descending camera flight across the shot. It placed countryside fields and buildings along the broad path of the original 3D camera.

When you plan a live-action shot this way, keep these details in mind:

• Pacing: The AI follows the general speed of your Blender camera across long takes.
• Broad paths: Sweeping landscape moves guide the overall terrain layout.
• Fine detail drift: Distant background details can shift slightly over 14 seconds.

2. Exhibition-installation previs (A)

Experiential designers and gallery artists often build forced-perspective art. In these setups, scattered physical pieces only align into a clear picture from one vantage point.

In our Scene A test, we placed broken shapes at different depths across a gallery room. We animated the camera moving sideways until frame 85. In our source file, the pieces line up into a butterfly shape at frame 85, and hold that alignment through frame 120.

Here's the five-second gray reference we built in Blender:

Scene A gray reference: camera moves sideways past floating shapes and holds an aligned butterfly shape from frame 85 to 120.

We submitted this clip as run 03-r1 with the prompt:

A quiet, human-scale art gallery measures roughly six by seven metres with clean rectangular doorframes and straight floor joints. Twelve thick, rounded sculptural pieces and one narrow central element rest stationary across three distinct depth layers in the room. Every sculptural element features a smooth finish of opaque white glazed ceramic, catching soft highlights along defined geometric contours. Restrained afternoon illumination enters from the side, casting clear, subtle shadows across the floor without heavy haze or obscuring blur. All sculptural components remain completely still in their fixed physical positions under even, balanced ambient daylight.

Here's the generated output:

Run 03-r1 result: the sideways camera track and final butterfly hold work well creatively, though floating pieces inflate into ceramic shapes during the move.

The test was an accepted creative success. The sideways track and final butterfly silhouette are clearly readable, and the hold at the end stays steady.

But the floating shapes don't stay as rigid geometric solids. As the camera moves, the thin shapes inflate and morph into glossy ceramic wings rather than acting like fixed physical fragments.

3. Miniature advertising and scale reveals (B)

Commercials for food, toys, or tabletop products often use scale reveals. A camera might fly through a tiny opening to show a whole miniature world inside.

In Scene B, we built a tabletop breakfast scene. The camera starts inside the hole of an upright cereal ring, backs out through the opening, and pulls back and rises to reveal a tiny explorer and miniature helicopter sitting in a cereal bowl.

Here's our gray reference video:

Scene B gray reference: camera backs out through a floating cereal ring hole, then pulls back and rises to reveal a tiny explorer and helicopter inside a milk bowl.

We submitted this reference in run 07 without any appearance image, using this text prompt:

A tabletop breakfast scene hosts an eighteen-centimetre plain glazed ceramic bowl containing still whole milk and six toasted cereal rings. One ring stands vertically in the pool, forming an open circular archway above the milky surface. Perched on the wide curved rim of the bowl, a three-centimetre painted helicopter sits grounded with distinct tail boom, landing skids, and unmoving rotor. Nearby, an eight-millimetre miniature pilot stands beside the aircraft, while an eight-millimetre stranded explorer waits on an adjacent cereal ring. Warm morning light illuminates the scene, catching linear grooves in the textured placemat below.

Here's the result:

Run 07 result: text prompt over Scene B creates rich milk textures and morning light as the camera backs out through the cereal ring and rises to reveal the bowl.

The output follows the camera backing out through the ring opening and revealing the bowl, rendering realistic milk, cereal textures, and morning light when you create a Seedance video.

The upright floating ring was already part of our Blender blockout, so the AI didn't invent that shape on its own. While the reveal works well, the exact size and position of the foreground ring change slightly during the move.

4. Character action beat

Directing multi-character fights with plain text prompts often causes tangled limbs and messy staging. Building simple proxy rigs in Blender lets you set body timing, kick extensions, and framing before generating video.

To check character action, we reviewed historical run ed811a27. This run used a 10-second, 1280x720 gray clay reference of two martial artists fighting.

Historical ed811a27 clip: Seedance 2.5, 480p, 10.04-second output from a 10-second 1280x720 clay reference. An extended front kick and bare feet remain visible, though framing is tighter and hit contact is soft.

The generated 10.04-second clip kept the extended front kick and showed the fighters' bare feet clearly.

The camera framing settled slightly tighter than our source clay pass. In addition, the physical contact and weight transfer during strikes showed slight visual softness.

Using 3D references for action helps you manage:

• Pacing and timing: Strikes land when your keyframed characters move in the source.
• Limb visibility: You can keep arms and legs from vanishing behind bodies.
• Staging: Both characters stay in their relative positions across the shot.
• Contact softness: Fast impacts will still look softer than hand-keyed visual effects.

5. Style exploration from one blocking file

Once you build a camera path in Blender, you can use that single 3D scene to test different art styles and materials. You don't need to re-keyframe your shot.

We tested this by passing Scene A through a second generation. Run 03-r1 used opaque white glazed ceramic materials. For run 04, we reused the exact same Scene A gray reference video, but changed the material text to polished copper.

Here's the prompt for run 04:

A quiet, human-scale art gallery measures roughly six by seven metres with clean rectangular doorframes and straight floor joints. Twelve thick, rounded sculptural pieces and one narrow central element rest stationary across three distinct depth layers in the room. Every sculptural element features a smooth finish of polished copper metal, catching soft highlights along defined geometric contours. Restrained afternoon illumination enters from the side, casting clear, subtle shadows across the floor without heavy haze or obscuring blur. All sculptural components remain completely still in their fixed physical positions under even, balanced ambient daylight.

Here's the result of run 04:

Run 04 result: polished copper test using the reused Scene A gray input; metallic reflections render across the forms while following the sideways camera path.

When you compare run 03-r1 and run 04, both clips follow the broad sideways tracking path and reach the butterfly silhouette.

Run 04 produced strong metallic reflections and warmer room highlights. The shapes still inflate slightly during the move, and lighting changes along with the material. But it shows how practical it is to test different looks when you keyframe your shot.

6. Game cinematic

Game trailers often feature frozen-time shots or dynamic camera moves around suspended action. Describing these complex frozen moments with text alone can be difficult.

In Scene C, we built a frozen street scene. A bicycle courier is suspended in mid-air above a vintage car, surrounded by floating papers. We animated a low camera push that rises laterally around the courier.

Here's our gray reference video:

Scene C gray reference: a low forward push rising laterally around a suspended bicycle courier and vintage car with loose papers in mid-air.

We submitted this reference in run 08 with the prompt:

A dramatic urban moment freezes completely over rough street asphalt marked by worn white pedestrian crossing stripes and three evenly spaced metal bollards. Suspended motionless above the hood of a stationary passenger car, a bicycle courier hangs mid-air alongside a lightweight bicycle, displaying one prominent foreground wheel. Six large delivery paper sheets remain locked at varying depths around the suspended cyclist, caught in static mid-flight. The courier wears a simple helmet, fitted jacket, and compact delivery bag. Grounded late-afternoon sunlight casts crisp directional shadows, preserving the intense physical tension while keeping all vehicles, debris, and figures strictly motionless.

Here's the generated output:

Run 08 result: warm urban atmosphere and lighting generated over Scene C; the broad camera arc and frozen subject layout survive without matching exact coordinates.

The output added brick buildings, golden-hour light, and car details. The frozen courier and floating papers stayed in place as the camera moved around them.

The move matched the general path of our Blender camera, though the generated elements don't match the exact 3D coordinates of the gray proxy shapes.

To explore more prompt ideas for stylized shots, check out our Seedance prompt examples.

What you can check in Blender before a paid render

Blender lets you catch framing and timing mistakes locally. You don't spend any AI credits when rendering local gray preview passes.

Here are five things you can check in your 3D viewport:

• Framing: Make sure your subject doesn't clip through the edge of the frame.
• Motion speed: Play back your animation at 24 fps to make sure the camera doesn't move too fast.
• Focal length: Lens perspective depends on camera distance. Changing focal length alters your field of view; to get telephoto compression, move the camera back and use a longer lens.
• Object shapes: Use clear, simple shapes so the AI can easily read what each object is.
• Timeline length: Make sure your reference clip is between 2 and 30 seconds to fit the add-on input limits.

For a full setup guide, see our seven-step blockout tutorial.

Which settings per use case

Different shot types need different reference setups. You can refine your camera terms using standard phrases from a practical prompt and camera vocabulary guide.

The table below shows the actual models, durations, reference focus, and limitations observed across our test cases:

If you want to dive deeper into camera paths, read our overview of camera controls and check the Blender camera add-on connection.

What this is not good for

Using a 3D blockout as a video reference gives you better framing control, but it has clear limits:

• No rigid geometry lock: The AI model doesn't freeze 3D meshes in place. In run 03-r1, floating shapes inflated and morphed during the camera move.
• Text can't easily cancel motion: In run 05, we added a prompt asking for a fixed camera over a moving reference. The camera still moved sideways. Conflicting text doesn't guarantee an override.
• No 3D camera data export: The add-on delivers an exported MP4 video, not 3D camera tracking curves or mesh files.
• No reference audio upload: The add-on accepts video references, not audio tracks. You can describe native sounds or foley in your text prompt, but you can't upload a reference audio track to sync against.

Knowing these limits helps you pick the right AI video tools for your production needs.

Open your own scene (CTA)

You can start testing your own camera moves in Blender today.

First, download the Blender add-on for Blender 4.2 or newer. Drag the ZIP file into your Blender window to install it.

Sign in once through your web browser, and you can run your tests directly from the Blender sidebar.

Build a simple five-second scene with basic shapes. Animate your camera, export a local gray pass, and write your prompt.

When you're ready to generate your shot, open the Seedance 2.5 video generator to turn your 3D blockout into a full video.

Related articles

• How to Turn a Blender Blockout into a Seedance 2.5 Video (7 Steps)
• Seedance 2.5 White Model Reference Prompt Guide
• How to Control Camera Movement in Seedance 2.5 with a 3D Camera Path