Landmark-Anchored Prompts: Camera Blocking That Actually Works
A quick 300-word overview on the essential techniques for landmark-anchored prompts: camera blocki...
Camera blocking in AI video generation has a fundamental problem: the model doesn't know where things are. When you write "the camera moves from the door to the window," the model has no understanding of where the door and window are in relation to each other, what's between them, or how the character relates to this space. The result is a camera movement that may have nothing to do with the spatial reality you described.
Landmark-anchored prompting solves this by defining spatial landmarks before describing any movement — building a mental map of the space that the model can reference when executing blocking instructions.
Why Standard Camera Blocking Fails
Consider a standard blocking prompt: the camera pans left to reveal the detective standing at the window
The model produces a pan and something that could be called a detective near something that could be called a window. But the spatial relationship is guesswork. Is the window behind the camera? To the left? Is the detective foreground or background? The output varies dramatically across generations.
Landmark anchoring gives the model a shared coordinate system — a set of named spatial positions that both the prompt writer and the model can reference consistently.
Step 1: Establish the Landmark Map
Before describing any camera movement, describe the spatial layout of the scene as a set of named landmarks.
Landmark map format: [Environment]: [Landmark A] is at [position in frame]. [Landmark B] is at [position in frame]. [Landmark C] is at [position in frame]. The camera starts at [starting position] facing [direction].
Example — interior office: Interior office scene: the door is at screen right, in the background. The desk is center-frame in the mid-ground. The window is at screen left, in the background. The camera starts at the doorway level, facing into the room.
With this map established, every subsequent blocking instruction has a shared coordinate system.
Step 2: Describe Movement Between Landmarks
Once the landmarks are established, describe camera and character movement in terms of those landmarks:
The camera begins at the door (screen right, background), pushes forward toward the desk (center, mid-ground), and settles in a medium shot as it arrives.
The character moves from behind the desk (center) toward the window (screen left), as the camera holds position — the character crosses the frame from center to left.
Compare this to: the person walks to the window
The landmark-anchored version produces a predictable spatial result. The unlabeled version produces whatever the model's default interpretation is.
Step 3: Establish Character Positions as Landmarks
Characters can also be named as landmarks once their position is established:
Maria is standing at the window (screen left). James enters through the door (screen right). The camera holds at the center, framing the space between them.
James walks from the door toward Maria, crossing the full width of the frame from right to left. Maria remains at the window, watching him approach.
The spatial relationship (Maria at screen left, James entering screen right, crossing to meet her) is now explicitly defined. The model has clear information about the spatial choreography.
Step 4: Camera Position as Landmark
The camera itself should be positioned as a landmark in complex blocking:
Camera A position: low angle at floor level, 3 meters from the desk, facing the desk. Camera B position: high angle above the door, looking down into the room.
For multi-shot scenes sharing the same environment, defining camera positions as named landmarks lets you describe coverage consistently:
From Camera B position: James enters, crosses to the desk, sits. From Camera A position: close-up on James's hands picking up the envelope.
Landmark Language Vocabulary
Screen positions:
- Screen left / Screen right
- Foreground / Mid-ground / Background
- Upper frame / Lower frame / Center frame
- Near camera / Far from camera
Physical positions:
- At the [named object]
- Behind the [named object]
- Between [Landmark A] and [Landmark B]
- Moving from [Landmark A] toward [Landmark B]
Camera positions:
- Camera starts at [position/landmark], facing [direction]
- Camera holds at [position]
- Camera tracks [subject] from [start landmark] to [end landmark]
- Camera pushes from [starting position] toward [target landmark]
Example: Full Landmark-Anchored Blocking Prompt
Scene: apartment bedroom, morning. Landmarks: the bed is center-frame in the mid-ground. The closet door is at screen left, in the background. The bathroom door is at screen right, background. The window is above and behind the bed. Camera starts at ground level at the foot of the bed, facing the headboard.
Maria wakes in the bed (center). She sits up, the window light hitting her from behind. She looks left toward the closet door (screen left). She gets up and moves left toward the closet, crossing the mid-ground from center to screen left. Camera holds its position at the foot of the bed, watching her cross the frame. She opens the closet door and looks inside. Cut.
This prompt produces predictable, directable output. The spatial logic is clear. The camera position and movement are explicit. The character's path across the frame is defined.
When to Use Landmark Anchoring
Landmark-anchored prompts are worth the setup time for:
- Any shot with character crossing or blocking
- Any camera movement that needs to start and end at specific positions
- Any scene where the spatial relationship between characters matters to the story
- Multi-shot scenes sharing the same environment
For simple shots (a character standing in one position, no significant movement), standard prompts are sufficient. The overhead of landmark description is only worth it when the spatial logic is essential to the shot.
Takeaways
- Standard camera blocking prompts fail because the model has no spatial coordinate system to reference
- Landmark-anchored prompts establish named spatial positions before describing any movement
- Define landmarks for environment features, character positions, and camera positions
- Describe blocking in terms of landmark-to-landmark movement rather than abstract direction
- Use the screen position vocabulary: screen left/right, foreground/background, near/far camera
- Apply landmark anchoring on any shot where spatial logic is essential to the story beat
Continue exploring
- Voice-First AI Directing: Director Studio, from Scene to Screen
- AI Scene Continuity: Inside the Persistent 3D Scene Rig
- The 180 Degree Rule in AI Video: Catch Axis Crossings Before Render
- Character Subtext in AI Video: Render the Tell, Not the Emotion Label