AI video models often stop following a prompt when one generation is asked to invent the visual style, character, location, composition, action, and camera movement at the same time. The instructions compete with one another, so important details drift or disappear.
The fix is not one enormous prompt. It is a visual system that can survive across every image, camera angle, and video generation.
For this project, I wanted to create a short film inspired by the unsettling visual language of the Backrooms: endless yellow rooms, flat fluorescent light, repeating architecture, and one person searching for an exit that never seems to appear.
The workflow has five parts:
- Analyze visual references and define one master style prompt.
- Turn the idea into a nine-shot storyboard.
- Place a consistent character inside the environment.
- Use sketches to control difficult poses and compositions.
- Animate the strongest frames with focused video prompts.
The key is to separate the project's permanent visual rules from the action of an individual shot. Once the style is locked, every new prompt only needs to explain what changes.
1. Turn Your References Into a Master Style Prompt
Start by gathering several frames that capture the world you want to build. Look for references that agree on the important visual decisions: color, lighting, lens, architecture, texture, scale, and mood.
Then ask Yapper Assistant to analyze those references as a group.
Agent Prompt: Extract the Visual Style
This produces a reusable style layer instead of a description tied to one frame. For the film, the final master prompt was:
Master Style Prompt for Images and Video
Use this prompt as the fixed visual foundation for both still images and video clips. The subject, action, room, and camera position can change, but the color grade, lighting, lens behavior, texture, and mood should remain stable.
2. Build the Film as a 3-by-3 Storyboard
Do not jump directly from an idea to video generation. First, ask the assistant for a storyboard that explores the story from several distances and angles.
Agent Prompt: Generate Nine Storyboard Frames
A useful nine-shot board might include:
- A wide establishing shot that makes the character feel small
- A medium shot of the character choosing between identical corridors
- A close-up of their confused reaction
- A low angle beneath the fluorescent ceiling panels
- A long one-point-perspective tracking composition
- A detail shot of wet carpet, damaged ceiling tiles, or abandoned furniture
- A side profile as the character listens for a sound
- A wide room that appears domestic but extends impossibly far
- A final frame that suggests the search will continue
The storyboard is not a commitment to use every frame. It is a fast way to test coverage before spending credits on animation.
Use Shot Size to Control the Feeling
Wide shots communicate scale and isolation. Medium shots make the search readable. Close-ups let the audience feel the character's anxiety.
If every shot is wide, the film can feel emotionally distant. If every shot is close, the environment loses its oppressive scale. A deliberate mix lets the edit move between the person and the impossible world surrounding them.
3. Keep the Character Consistent Inside the Environment
Once the storyboard works, bring in a clear character reference. Keep the character's identity, glasses, outfit, and proportions stable while changing only their movement and camera position.
Agent Prompt: Place the Character in the Scene
The result gives you several candidates with the same narrative purpose. Choose the frame with the clearest silhouette, strongest corridor geometry, and most useful direction of movement.
For more control, describe the character and the room separately, then reapply the master style. This keeps the model from replacing the environment when you ask for a new pose.
Example Still Prompt: Searching the Yellow Hallway
This frame was generated with Nano Banana Pro.
Notice that the shot prompt does not replace the master style. It translates that style into a specific composition, expression, wardrobe, and camera height.
4. Use a Sketch When Words Are Not Precise Enough
Some shots are easier to draw than describe. A rough sketch can communicate body angle, arm position, silhouette, framing, and the direction a character should face.
You do not need to make polished concept art. A simple drawing can act as a composition map while the character reference controls identity and the master prompt controls style.
Agent Prompt: Turn a Sketch Into a Finished Frame
When you submit the request, give each reference one job:
- The character image controls identity, wardrobe, and proportions.
- The sketch controls pose and composition.
- The master style prompt controls color, lighting, texture, lens, and mood.
This division of responsibility is more reliable than asking one reference to control everything.
5. Animate the Best Frames With Focused Motion
After you have a strong still frame, animate only the movement that matters. Preserve the world and character, then specify the action, camera behavior, and emotional beat.
The following examples were generated with Seedance 2.5.
Example Video Prompt: The Endless Yellow Kitchen
This prompt works because it gives the clip one clear physical sequence. The environment is impossible, but the action—putting down a knife, lifting a board, walking toward the stove—remains simple and observable.
Example Video Prompt: Liminal Bedroom Dread
The explicit hard cut separates two clear beats: preparing to sleep, then realizing sleep is impossible. The instruction to avoid additional cuts prevents the model from inventing coverage that weakens the moment.
A Reliable Prompt Structure for Every Shot
For each new generation, keep the prompt in the same order:
- Subject: Who is in the frame, what they are wearing, and what they do.
- Scene: The specific version of the liminal environment.
- Composition or camera: Shot size, angle, lens, and movement.
- Lighting: The fluorescent ambient rules from the master style.
- Style: Color grade, texture, haze, distortion, and found-footage treatment.
- Mood: The emotion the shot should create.
- Constraints: Anything that must not change, including cuts, identity, wardrobe, or props.
Repeating this hierarchy makes prompts easier to debug. If a result fails, you can usually identify whether the problem belongs to the subject, scene, camera, style, or motion instead of rewriting everything.
Edit the Film Around Escalation
The finished sequence should become more unsettling without relying on a monster or a sudden violent reveal.
Begin with exploration. Let the audience understand the repeating corridors. Then introduce impossible variations: a kitchen that never ends, a bedroom with no privacy, furniture abandoned in the middle of a vast office-basement space. Finish with the character unable to rest or find an exit.
In the edit:
- Alternate wide shots with closer emotional coverage.
- Keep the sickly yellow grade consistent across every clip.
- Use fluorescent hum, distant room tone, footsteps, carpet movement, and appliance sounds to make the space feel real.
- Let shots run long enough for the audience to inspect the repetition.
- Avoid overusing glitches, jump scares, and digital effects. The architecture should create the dread.
Final Checklist
Before exporting the film, confirm that:
- The same character identity and wardrobe survive every shot.
- The walls, carpet, ceiling tiles, and fluorescent fixtures share one yellow-beige palette.
- Light stays flat and ambient instead of becoming dramatically cinematic.
- Wide shots preserve deep, symmetrical vanishing lines.
- Lens distortion, haze, and grain feel consistent rather than random.
- Every clip has one readable action and no unnecessary camera cuts.
- The sequence moves from curiosity to isolation and quiet dread.
The master prompt is what holds the project together. The storyboard gives it cinematic coverage, the references preserve identity and pose, and the video prompts add only the motion needed for each beat. That combination turns a collection of AI generations into one coherent short film.