Driving Blender and After Effects by Prompt

Most builders write off 3D and motion graphics as out of reach. Blender has a thousand-button UI and a node graph that punishes you for blinking. After Effects hides its real power behind expressions most people never learn. So the median product ships flat — a static hero image, a CSS fade, maybe a Lottie file someone bought. That's the motion equivalent of the purple-gradient tell: it reads as "we didn't have anyone who could do this."

You don't need to learn the thousand buttons. You need to direct the tool while Claude operates it. Over MCP, Claude writes the Blender bpy Python — keyframes, geometry nodes, materials, render setup — and the After Effects ExtendScript — expressions, effects, comps. You describe the shot in plain language; the AI handles the scene graph. This is the same move as overwriting the agent's default taste with a skill, applied to a medium most developers skipped entirely. The pillar piece argues the model ships the median; here the median is no motion at all, and beating it is now a prompt.

Key Takeaways

• The scene graph is the barrier, not your taste. Claude operates Blender and After Effects programmatically over MCP, so you direct the shot instead of memorizing menus.
• Two skills, two pipelines. blender-motion handles 3D, geometry, and render; aftereffects-motion handles 2D motion graphics, expressions, and compositing — install only the one your shot needs.
• Each skill needs its own MCP bridge. Blender needs the Blender MCP add-on running; After Effects needs the AE MCP. The skill writes the script; the bridge runs it in the app.
• Specify the feel, not the keyframes. "Logo scales in with overshoot, settles by frame 24" gives the AI everything it needs to write the easing — you art-direct, it implements.
• This is the deep design layer. Motion is part of the Designs category discipline, not a separate craft you outsource.

Why Text Beats the UI Here

Blender and After Effects are both, underneath, scriptable scene graphs. Blender exposes bpy, a full Python API over every object, modifier, keyframe, and render setting. After Effects exposes ExtendScript (JSX), which reaches every layer, property, and expression. The GUI is just one front-end to those APIs — and it's the slow one for anyone who isn't a daily user.

When Claude drives the API directly, the friction that kept you out disappears. You don't hunt for the keyframe interpolation dropdown; you say "ease out, slight overshoot" and the AI writes a Bezier handle or a transform expression that does exactly that. The model already knows the bpy and JSX surface cold. Your job collapses to the part that was always the real work: deciding what the shot should feel like. That's art direction, and it's the part a skill can't do for you — see trust is a design material for why taste stays with the builder.

Blender vs After Effects: Which Skill

The two tools solve different problems, and so do the two skills. Reach for Blender when the shot is genuinely three-dimensional — a product turntable, a camera move through geometry, real lighting and materials. Reach for After Effects when the shot is 2D motion graphics — animated logos, kinetic type, UI mockups in motion, compositing.

Install only what the shot needs. A turntable hero wants Blender; an animated logo for a landing page wants After Effects. Both assume their MCP bridge is running — the skill writes the script, the bridge executes it inside the live app so you see the result and can iterate.

A Turntable in Blender

Say you want a clean product turntable — the kind of looping 360° spin that makes a hero section feel like a real product launched it. With the Blender MCP add-on running and the skill installed, you direct it like this:

Build a 5-second perfume-bottle turntable. Import the bottle mesh,
center it on the world origin, and parent it to an empty. Animate the
empty rotating 360° on Z over 120 frames, linear, looping clean.
Three-point studio lighting, soft key from front-left, subtle rim from
behind. Add a reflective floor with a slight gradient falloff. Camera
at 35mm, slightly above eye level, locked. Set the world to a dark
neutral. Render Cycles, 1080p, 24fps, transparent background.

Claude turns that into bpy — it sets the rotation keyframes on the empty, builds the lamps, assigns the floor material, configures the camera focal length, and writes the render settings. You watch it appear, then refine in the same plain language: "kill the rim light," "slow the spin to 6 seconds," "make the floor darker." Every correction is a sentence, not a hunt through the properties panel. The bpy-aware loop is exactly the kind of behavior covered in designing AI product behavior — the AI handles repair and iteration so you stay in the director's chair.

A Logo Scale-In With Overshoot in After Effects

Now the 2D case. You want a logo that scales in with a touch of overshoot — the small bounce that separates "designed" motion from a lazy linear fade. With the AE MCP running:

In the active comp, animate the logo layer scaling from 0% to 100%
over 24 frames. Add overshoot: it should slightly overshoot to ~108%
around frame 18, then settle to 100% by frame 24. Use an expression
so the easing is editable, not baked keyframes. Add a 6-frame opacity
fade-in on the same layer. Then drive a soft drop shadow that grows as
the logo settles.

Claude writes the ExtendScript that adds the layer animation and, crucially, attaches an expression for the overshoot rather than hard keyframes — so the bounce stays parametric and you can tune it. Ask for "more bounce" and it edits the expression's amplitude; ask for "snappier" and it shortens the settle. This is the same parameterized-template thinking from system prompts are design, applied to motion: expose the dial, don't bake the value.

Frequently Asked Questions

Do I need to know Blender or After Effects to use these?

No — that's the point. You need to know what you want the shot to feel like. Claude writes the bpy Python or the ExtendScript; you art-direct in plain language. Knowing the apps helps you give sharper notes ("ease out," "rim light," "parametric overshoot"), but you don't need to operate the UI yourself.

What's the MCP bridge for?

The skill writes the script, but something has to run it inside the live application. The Blender MCP add-on and the After Effects MCP are those bridges — they execute Claude's generated code in the running app so you see the result immediately and iterate. Without the bridge, you'd have a script with nowhere to run it.

Can I use both skills on the same project?

Yes, and you often should. A common pipeline renders a 3D element in Blender, then composites and adds 2D motion-graphics polish in After Effects. Install both, run both MCP bridges, and let Claude hand work between them. Keep your direction specific to whichever stage you're in.

Why not just generate a video with an AI video model?

Generative video gives you a clip you can't edit precisely — you can't say "change the overshoot to 108% and settle two frames earlier." Driving Blender and After Effects gives you a real, editable project file with named layers, keyframes, and expressions. You get art-direction control, not a one-shot render you're stuck with.

Where does this fit in the anti-slop stack?

Motion is a design surface most products leave blank, which makes it high-leverage. It sits alongside the other deep layers in the six-layer anti-slop stack — pixels, behavior, prompts, trust, and motion. Beating the median here is cheap now and still rare, which is exactly where you want to spend effort.

Explore motion and the deeper design layers in the Designs category, or browse the full skill catalog at aiskill.market.