Works with the AI Agents you already use
madcam
Engineers impossible camera prompts and surreal narrative logic for high-end AI video models.
Free
madcam
Example session with this skill installed
I have @Image1 (a neon-lit Tokyo alley) and @Image2 (a glass marble). Make a 10s madcam sequence where the marble is the protagonist. 5 shots. Hard cuts.
- Read your context and instructions
- Compiled the madcam
"THE GLASS EYE" — 10s, 5 shots x 2s, hardcut chain
Logline: A glass marble rolls through a neon alley, and wherever it touches, the city's electricity is pulled inside it.
Rule of the World: Glass acts as a vacuum for light.
References
- @Image1: Primary Environment (Tokyo Alley)
- @Image2: Protagonist (The Marble)
Full Prompt (Paste-ready)
Global: Cinematic photorealism, 16:9, 24fps. Hard cuts only. No motion blur or camera shake. Lighting: High-contrast cyan/magenta neon. Identity lock: Marble texture from @Image2, Alley architecture from @Image1.
SHOT 1 (0–2s): THE UNDER-SOIL EYE
Camera: Inside the pavement looking up through a translucent glass crack.
Move: Static.
Subject action: The marble (@Image2) rolls directly over the crack; as it passes, the neon light from the alley walls visibly bends and gets sucked into the marble's core.
Impossible element: Light behaves like a liquid being vacuumed.
Light: Bottom-lit magenta glow from the marble.
Cut-out: Circular shape of the marble center-frame.
SHOT 2 (2–4s): THE PUPIL AXIS
Camera: 360-degree orbit locked on the center of the marble.
Move: High-speed rotation.
Subject action: The alley (@Image1) spins into a blur while the marble remains perfectly still and sharp in the center; sparks of blue electricity jump from wet pavement to the glass.
Impossible element: Environment rotates while the rolling object stays stationary in screen-space.
Light: Flashing cyan strobes.
Cut-out: High-speed rotational motion.
SHOT 3 (4–6s): THE REFLECTION TUNNEL
Camera: Lens lives inside the reflection on the marble's surface.
Move: Zooming "out" from the reflection to reveal the real world.
Subject action: We see a distorted Tokyo tower in the glass; the marble hits a puddle, and the reflection and the real world swap places.
Impossible element: Reflection becomes the physical reality.
Light: Golden hour sun hitting the glass edge.
Cut-out: Water ripple texture.
SHOT 4 (6–8s): THE MACRO LADDER
Camera: 5mm Macro lens.
Move: Tracking millimetres behind the marble.
Subject action: The marble rolls over a discarded circuit board; the traces on the board light up and liquefy, flowing into the glass.
Impossible element: Solid metal traces turning into liquid light.
Light: Rim light on the glass edge.
Cut-out: Bright white flare.
SHOT 5 (8–10s): THE LIGHT-RAY POV
Camera: Lens is a beam of light trapped inside the marble.
Move: Forward thrust.
Subject action: The marble stops suddenly; all the collected neon light erupts forward like a laser, illuminating a dark doorway.
Impossible element: A marble acting as a high-intensity projector.
Light: Absolute darkness except for the forward beam.
Next steps
- Generate each 2s segment separately for maximum control over the "Impossible element."
- Use the "Reflection Tunnel" shot as a prompt for a dedicated image-to-video seed.
- Request a "Split-screen version" to show the light vacuum from multiple angles.
Connects securely to your tools. The creator never sees your data.
About this skill
Standard video prompts often rely on generic drone shots and flat cinematography. This skill forces AI video models like Kling, Veo, or Sora to generate impossible, physics-defying camera work paired with tight narrative logic. It moves beyond "cool visuals" by anchoring every camera choice in a specific story rule.
What it does
- Impossible camera design creates shots that cannot exist in reality, such as lenses inside marble, pupil-axis orbits, or macro-to-planetary zooms.
- Narrative engine building defines a single visual "Rule of the World" that drives the logic of the entire sequence.
- Technical shot templating provides specific optics, focal lengths, and lighting directions that video models actually recognize.
- Advanced editing modes structures content as match-cut chains, complex sequence shots with host-relay logic, or geometric split screens.
- Model-safety auditing filters out prompts likely to cause morphing, identity loss, or physics glitches in the AI generation process.
How it works
- Input mapping identifies your character, location, and prop references and binds them to specific roles.
- Rule definition establishes the impossible visual rule, such as shadows arriving before the person or sound leaving scars.
- Shot assembly selects cameras from specialized "armories" like Matter, Body, or Space-Warp families.
- Prompt delivery outputs a paste-ready global configuration followed by time-coded shot blocks.
Frameworks & tools
Compatible with high-end AI video generators including Kling, Luma Dream Machine, Runway Gen-3, Kling, Seedance, Higgsfield, and Sora.
Why this beats prompting it yourself
Manual prompting often results in "floating camera" syndrome where the lens moves without purpose. This skill uses a structured 6-knob system (Where, What Moves, Lens, Scale, Time, Light) to ensure every frame has high-end cinematic intent. It prevents the model from defaulting to clichés like generic orbits or sparkles.
Use cases
- Concept trailers for sci-fi or fantasy films needing a unique visual signature.
- High-end commercials where a product must interact with the world in a surreal way.
- Music video treatments that require rhythmic, impossible transitions and split-screen geometries.
- Social media "hooks" designed to stop the scroll with imagery the viewer has never seen.
Known limitations
The skill focuses on cinematic structure and does not generate actual video files or images. Performance depends on the specific video model's ability to interpret complex spatial instructions.
Requires a video generation model that supports detailed prompting. Best results with models that can handle specific camera/lens terminology. Output is text-only instructions.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account