Conditions for using Blender and 3D for video generation: designing cameras, occlusions, and continuity

3D is not a tool to improve quality, but a tool to reduce ambiguity

In the generated video, scenes such as a person walking around a table,'' two people exchanging glances,'' and ``a product being pushed toward the front'' are the only scenes that break down many times. Before switching to an expensive model, there is one question that remains for the production side. Can you explain the position of the camera and objects yourself? A space that cannot be explained will not be stably transmitted to the model.

You should use Blender not only when you need a photorealistic final render. This is when you want to fix things that can be hidden within the screen, people's movement paths, light sources, camera parallax, and the 180 degree rule between multiple cuts. Even blocking with only gray boxes and stick figures has value. In fact, if you create the decoration first, it is easy to mistake a spatial problem for an art problem.

1. 1Script Movement
2. 2Place Floor/Wall/Desk in Box
3. 3Person Position

1. 1Standing Position
2. 2Camera and Lens
3. 3Rough Render/Depth/Contour

1. 1Rough assets
2. 2reference image or control
3. 3video suggestions
4. 4edit to match

Hold the camera with both words and numbers

The angle of view of the camera changes the meaning of the subject. Wide angle makes nearby hands larger and exaggerates the distance between foreground and background. A long focal length compresses perspective and makes the background appear closer. "It looks like 50mm" is useful common knowledge, but the actual appearance depends on the sensor size, distance to the subject, and cropping. Refer to Blender's camera documentation and write not only the focal length but also the camera height, subject distance, orientation, and start and end points of movement in the production notes.

Shot cards do not stop at "close" or "cinema style". For example, set your eye level to 1.55m, 2m from the subject, look over your shoulder, move slowly 25cm forward, and leave the window in the background one-third to the left. With this description, it can be inspected in 3D and only the necessary parts can be transferred to the video model prompt. Although there is no guarantee that the generator will reproduce the exact physical lens, fixing the intent allows for selection of candidates.

Five stages of blocking

Place the floor first and decide on north. Second, place walls, windows, desks, and doors with boxes. Third, place the person as a sphere and a capsule, and mark the start and end positions of each cut on the frame numbers. Fourth, place the main light, auxiliary light, and window light in only the general direction. Fifth, place the camera and check the line of sight, direction, and brightness that the next cut will receive on the last screen of the cut. A few seconds of low-resolution previews are sufficient at this stage for untested design exercises.

1. 1Determine spatial coordinates
2. 2Person A goes from left to right on the screen

1. 1Exit of cut A: right gaze
2. 2entrance of cut B: place A on the left and continue to the right

1. 1Beyond the cut axis
2. 2Insert neutral shot
3. 3Teach the audience a new axis

The 180 degree rule is not a law. The editorial promise is that by keeping the camera on one side of the line connecting the two people, it will be easier for the audience to maintain left-right relationship. If you cross the line, include a frontal neutral shot, a clear movement shot, and a point of view shot to signal your new position. A generated video may look beautiful in just one shot, but if the person's shoulders or direction of movement are reversed in the next shot, it feels strange. Confirmation by arranging the final frames as still images is inexpensive and highly effective.

Situations where it is better not to use 3D

The strict design of 3D can slow exploration in images that appeal to emotional leanings, abstract transformations, shapes like smoke or watercolor, and randomness. If the cost of creating a 3D asset is greater than several generations of generation and selection, a reference photo or rough sketch will suffice. Part of speed is deciding not to open Blender if there is only one final cut and there is no occlusion or contact between multiple people.

On the other hand, if there are product advertisements, architecture, conversations between the same characters, movement of vehicles/machines, movement of multiple cuts, or camera matches, coarse 3D will reduce the reproduction cost. LTX-Video officially guides the control paths of Pose, Depth, and Canny, but check separately in which distribution weight, UI, and version these can be used before execution. The control image is not a completed image, but a blueprint that conveys the composition and posture that you want the generator to follow.

Continuity check

Check the following for each cut. These are the left and right side of the screen, the line of sight of the person, the direction of movement, clothing and props, objects held in hand, direction of light, openings in the background, and direction of sound. If something moves from the right hand to the left hand, either create a cut where you can see the movement, or create a composition that allows what happened outside of the cut to be naturally accepted. Simply asking the generator to "be consistent" leaves it ambiguous as to what to compare.

3D blocking deliverables are not just .blend files. It is a rough PNG of each shot, a camera chart, a chart of the positions of people and objects, and a contact sheet that lists the last and the next beginning. If you have these four things, you can pass on your intentions to another model or another editor. Once the space is clearly defined, there is a clear distinction between what will be replaced by advances in the model and what will remain as part of the video production process.

20 minute blocking procedure in Blender

Set the unit to meters, set the frame rate to the value you plan to edit, and place the floor. Walls, windows, desks, and chairs can be boxes that prioritize positional relationships over dimensions. The person is made into a temporary doll with the head, chest, and pelvis as a sphere, and the arms and legs as cylinders, and markers are placed at the beginning and end points of the cut. Make the props a box of the same height and width. The goal is not to look good, but to know what's in front of what.

The camera starts at line-of-sight height and checks eyes, hands, products, and windows within the screen. After changing the focal length, adjust the camera distance to match the subject size. Place keyframes at the start and end points and preview the background parallax and window movement. The role of control passed to the generator is also limited to one, such as depth is front and back, outline is composition, and pose is skeleton. We do not leave the details of art to the rough.

1. 1Place floors, walls, and desks in boxes
2. 2Starting position of people and props

1. 1End position and camera exit
2. 2place next cut entrance

1. 1Contact Sheet
2. 2Match Left/Right/Gaze/Light/Object
3. 3Output Rough for Generation

Diagnosis order when fixing continuity failures

When you feel that "the room is different" after generation, do not make the prompt longer immediately. First, classify what information in the rough has been lost. If the front and back of the desk and window are reversed, it is the depth or camera position; if the direction of the person's movement is reversed, it is the axis or entrance/exit frame; if the product size is different, it is the reference object in the screen; if the face orientation is different, it is the pose or line of sight reference; if the light is strange, it is the direction of the main light. If you try to fix these five things at the same time in one regeneration, you won't be able to observe the cause.

As a specific example, consider three cuts in which a person walks from the right side of a desk to the left and stops in front of a window. In the first cut, the person is placed on the left side of the screen and moved to the right. If the camera turns to the opposite side in the second cut, either show the movement itself on screen, or insert a neutral shot from the front. When switching to near the window in the third cut, the line of sight and the direction of the window light are continuous, rather than which side of the screen the person is on. The 3D preview is not a ``correct image,'' but a picture-story show that allows you to find out this selection in advance.

Use checklists at the beginning and end of each cut. Which direction does the person face? In which hand are things held? Where does the principal light come from? Where are the doors and windows in the background? Which side of the axis is the camera on? Where does the sound come from? If there is even one question that cannot be answered, fill in the uncertain conditions of the shot design before suspecting the instability of the generator. In this way, the earlier the spatial judgment is completed, the faster the evaluation of generation candidates will be.

What 3D still contributes after controllable generation

From 2024 onward, product announcements increasingly described camera and reference controls, but neither replaces an explicit spatial plan. 3D blocking remains valuable because it gives a reviewable answer to where the camera, subject, occluder, and light are before generation. Save a low-resolution turntable or viewport capture with the lens, camera height, and target marker. It lets a collaborator diagnose a flipped axis or impossible screen direction without treating generated pixels as ground truth.

No primary-source video-control release specific to this chapter was confirmed in the 2026-09-04 to 2026-10-04 boundary. That absence is a reason to test the currently available tool, not to infer that control has stopped improving.

HDR makes the camera plan extend into compositing

Runway's September 23 HDR guide explains HDR as a delivery concern involving precision, color space and display behavior; it is not a guarantee that a generated clip is correctly mastered. When 3D blocking supplies depth or a rough render, record the working color space, display transform, reference display, and where the image is converted for export. A camera match judged on an SDR preview can fail after an HDR transform.

Exercise: block one shot in Blender, export a neutral reference frame, and make an SDR review export plus a separate HDR candidate where the chosen tool supports it. Compare the same boundary frame on the intended displays. Diagnose geometry, exposure, and transform separately; do not cure a camera mismatch by grading it away.