Mastering Multi-Part Prompts

From Blueprint to Command: The Anatomy of a Prompt

A common mistake when starting with AI is to write a short, descriptive sentence and hope for the best. To achieve the control needed for storytelling, we need to think more like a programmer or a technical director. A powerful prompt is not a single request; it's a structured set of instructions.

To give our prompts a robust and predictable structure, we'll adopt a five-part formula that works well across many AI video models.

As you can see, this [Camera] + [Subject] + [Action] + [Setting] + [Style & Audio] formula is a perfect match for the information we've already gathered in our Shot Spec Cards.

Translating Your 'Shot Spec Card' into a Prompt

Let's take the example from our last lesson for the first shot of our story about the robot, Bolt.

Shot Spec Card (Recap):

• Cinematography: Medium Shot, slow dolly in.
• Action: Bolt walks into frame, stops abruptly. His head tilts down slowly, and a small blue light on his antenna begins to blink with curiosity.
• Style & Ambiance: Inspired by 'WALL-E'. Shot as if on Panavision Millennium DXL2 with Zeiss Prime lens. Gritty, dusty atmosphere. Late afternoon golden light.
• Audio: SFX: soft 'whirr-click' sound. Ambient: distant wind howling.

Now, let's combine these elements into a single, structured text prompt. We simply string the clauses together, often using commas or periods to separate the distinct ideas.

Resulting Prompt:

Medium shot, slow dolly in. A small, lonely robot named Bolt walks into frame and stops abruptly in a junkyard aisle. His head tilts down slowly as a small blue light on his antenna begins to blink with curiosity. The setting is a gritty, dusty junkyard in the late afternoon golden light. Style is cinematic, inspired by WALL-E, shot as if on a Panavision Millennium DXL2 with a Zeiss Prime lens. Audio: SFX of a soft 'whirr-click', ambient distant wind howling.

Refining the Components for Maximum Control

A basic translation is a great start. Now, let's see how we can make each part of the prompt even more powerful.

1. Specifying the Subject (Character)

To maintain character consistency—a major challenge in AI video—your description of the subject needs to be incredibly detailed and consistent in every prompt.

For our robot Bolt, instead of just "a small, lonely robot," we might create a detailed block:

Subject: A small, weathered robot named Bolt, about 2 feet tall. His chassis is made of dented, rusted orange metal plates. He has large, expressive binocular-style eyes that glow a soft blue. He moves on two worn-out caterpillar tracks. A single, thin antenna protrudes from his head.

You would then include this entire block in every prompt related to Bolt.

2. Specifying Action and Camera (Technical Direction)

Modern AI models are becoming less like image finders and more like physics simulators. This means we can direct them using technical language about forces, physics, and camera movements.

Using the "Causal Chain" technique for Sora, an action like "a cup falls" becomes:

"A stray elbow knocks a cup. Causality: The cup tips on its fulcrum, liquid sloshes, and the impact force shatters the glass."

This gives the AI logical steps to simulate, resulting in much more realistic motion.

Advanced Structures for Veo and Sora

Since you have access to Veo and Sora, you can use prompting structures specifically designed for their capabilities. These are the ultimate form of multi-part prompting.

The Veo JSON Schema is particularly powerful. It forces you to be specific about every single element, from camera lens to lighting contrast, preventing "concept bleed" where describing the mood accidentally changes the subject.

{
  "project": "Bolt's Discovery",
  "aspect_ratio": "16:9",
  "cinematography": {
    "shot_type": "Extreme Close-up",
    "movement": "Slow rack focus from foreground to background",
    "lens": "Zeiss Prime, f/1.4",
    "focus": "Starts on sprout, ends on pot rim"
  },
  "subject": {
    "description": "A single, tiny green sprout with two leaves, growing from cracked earth",
    "action": "A water droplet lands on one leaf, causing it to tremble slightly"
  },
  "environment": {
    "location": "Inside an old, rusty cooking pot in a junkyard",
    "lighting": "Late afternoon golden light, high contrast"
  },
  "style": {
    "aesthetic": "Cinematic, hopeful but gritty",
    "color_palette": "Vibrant green, rusty orange, dusty brown"
  },
  "audio": {
    "sfx": "Single, clear water drip sound",
    "music": "A faint, gentle musical note begins to swell"
  }
}

Conclusion

You have now learned to translate your creative vision from a planner's blueprint into a director's command. By structuring your prompts with deliberate, multi-part detail, you gain significant control over the AI's output, laying the foundation for a coherent narrative.

Key Takeaways:

• Effective prompts are structured and multi-part, not a single vague sentence.
• The [Camera] + [Subject] + [Action] + [Setting] + [Style & Audio] formula provides a reliable universal framework.
• Extreme detail in your Subject description is a key strategy for character consistency.
• Modern AIs respond to technical direction, such as specific camera movements and physics-based "causal chains."
• Tools like Veo and Sora offer advanced, structured prompting formats (like JSON) that provide maximum control.

We've focused on writing a detailed prompt for a single shot. But how do we guarantee that Bolt looks identical from the shot where he discovers the plant to the next shot where he interacts with it?

In our next lesson, we will tackle this challenge head-on. We'll learn how to use anchor images and character reference sheets to maintain visual consistency across multiple clips, ensuring your story's world and its inhabitants remain stable and believable.