After Effects extension · AI-native startup
Physics for your layers. In 2D and 3D.
Gravity simulates rigid bodies on your After Effects layers, lets you preview right in a docked panel, then bakes the result into ordinary keyframes on a duplicate comp. Built by an AI agent team: design, code, tests and QA.
Why Gravity
Believable physics shouldn't take all afternoon.
A bounce, a stack, a logo that tumbles into place. By hand, each one is a pile of keys, timing by eye and redo after redo. Gravity lets the physics engine work out the motion and leaves you with keys you can still art-direct.
The hard way
Hand-keyed physics
- Every bounce, spin and settle is keyed and timed by eye.
- Change one object and the rest of the chain needs re-timing.
- 3D tumbles mean fighting rotation curves and Orientation flips.
With Gravity
Simulate, preview, bake
- Turn the layers you already have into bodies and press play in the panel.
- Change friction, bounce or gravity and preview again.
- Bake to plain keyframes on a duplicate comp, then fine-tune by hand.
Workflow
Four steps. Never leave After Effects.
-
01
Rescan
Gravity reads the layers, cameras and work area of the active comp.
-
02
Select bodies
Choose Static, Dynamic or Kinematic per layer, in AE layer order.
-
03
Preview
Play to cache the simulation and scrub any frame.
-
04
Bake
Write keyframes to a duplicate comp.
Objects in these cards are illustrative renders, not Gravity output.
3D
Real depth. True Z.
3D layers move in X, Y and Z and tumble freely. Rotation bakes to X/Y/Z Rotation rather than Orientation, with multi-turn handling. Mix 2D and 3D layers in the same comp.

-
Look through your AE camera, or orbit the scene.
-
OBJ and glTF models collide as a convex hull, with a box fallback you're told about.
-
Point gravity anywhere, including -Z and +Z.

Preview = Bake
What you preview is what you bake.
Bake reads the exact frames you cached. It doesn't run the simulation again. Bake a whole comp, the work area, or just a range, and update only that range in a previous bake.











Curves
Keys you can edit by hand.
Editable curves are on by default. Gravity keeps keys only where motion means something and joins them with regular Bezier keys.











- Collision - split tangents, so the bounce angle reads.
- Apex - smooth turn at the top of a bounce.
- Rest - flat tangents, no keys while still.
- Correction - keys where needed to meet tolerance.
For example, an object that bounces three times over 150 frames usually lands around 10 keys.
Tolerance defaults: 0.5 px position, 0.1° rotation. Turn it off for a key on every frame.
Safe
Your original stays yours.
- Bakes always go to a duplicate composition.
- One After Effects Undo removes the bake.
- Plain keyframes. Whoever opens the project doesn't need Gravity.
For important projects, save a copy first.

Features
Depth when you need it.

Joints
Fixed, Pivot, Piston, Spherical, Spring and Rope.

Kinematic release
Animate with keys, then hand off to physics after the last key or at a frame.

Split text into letters
One shape layer per letter, each colliding on its own outline. For a supported 2D text layer.

Shape-aware colliders
Shape paths, masks, stars and ellipses collide on their outlines.

Collision groups
Eight groups, A to H, with a matrix view.

Keyframe the physics
Animate friction, bounce, damping and gravity over time.

Simulation bounds
Comp, 2x, Custom or Unlimited, shown as a dashed box.

Direct handles
Drag start pose, velocity, center of mass and gravity right in the viewport.
Objects in these cards are illustrative renders, not Gravity output.
Feature detail
Everything in the panel.
The full list, as the panel ships in the development build. Nothing here is a roadmap item.
- 01Bodies
- Static, Dynamic and Kinematic per layer
- Kinematic layers follow their AE keys, then release to physics after the last key or at a chosen frame
- 2D and 3D layers in one simulation world
- 02Collision shapes
- Shape layer paths and masks collide on their outlines
- Parametric rectangles, ellipses and stars
- Text and footage use their bounds; split text into letters for per-glyph outlines
- OBJ and glTF models as a single convex hull, with a visible box fallback
- 03Materials
- Friction, bounce (restitution), damping and gravity scale per body
- All of them keyframeable, with a graph and a table editor
- Outcome presets: Soft drop, Bouncy, Heavy and sticky
- 04Joints
- Fixed, Pivot, Piston, Spring and Rope
- Spherical joints between two 3D bodies
- 05World
- Gravity direction on X, Y and Z, with Down, Up, Left, Right, -Z and +Z presets
- Simulation bounds: Comp, 2x, Custom (six faces) or Unlimited
- Collision groups A to H with a matrix
- 06Viewport
- Comp view, AE Camera (matches the active camera) and Orbit with Front, Right and Top
- Drag handles for start pose, velocity, anchor and center of mass, and gravity; Esc cancels
- 07Timeline and bake
- Work area, cache range, bake range and physics keys in one strip, with a stale-cache warning
- Bake the comp, the work area or a custom range; update only that range in a previous bake
- Editable curves with 0.5 px / 0.1° default tolerances, or a key on every frame
- 08Safety and privacy
- Every bake goes to a duplicate comp; one AE Undo removes it
- Clear reasons when a layer can't be a body
- Diagnostics are a local ZIP you create; nothing is sent automatically
Use cases
Made for motion design work.
Motion graphics
Drop, stack and knock over shape layers that collide on their real outlines, then keep the curves for timing passes.
Shape-aware colliders · Editable curves
Kinetic typography
Split a supported 2D text layer into letters and let every glyph fall and collide on its own outline.
Split text into letters · Collision groups
Logo animation
Animate the logo with keys, then let it drop into place with physics after the last key or at the frame you choose.
Kinematic release · Joints
Product and 3D shots
Tumble 3D layers and OBJ or glTF models through your AE camera, with rotation baked to X/Y/Z Rotation.
3D layers · Convex hull models · AE Camera
Objects in these cards are illustrative renders, not Gravity output.
Compare
How it differs from Newton 4.
| Gravity | Newton 4 | |
|---|---|---|
| Simulation space | 2D and XYZ 3D rigid bodies | 2D rigid-body simulation |
| Preview to output | Bake reads the preview frame cache | Render solves the simulation again before converting to keys |
| After Effects outputCommon ground | Editable keyframes on a duplicate comp | Editable keyframes, with an option to apply to a copy |
Based on each product's public documentation, checked 2026-09-29. Newton is a trademark of Motion Boutique. Newton can still be used to create 3D-looking shots.
Built with AI · AI-native startup
Built by an AI agent team. Solved by a real physics engine.
We are a small AI-native startup. An AI agent team worked with us on the design, code, tests and QA of Gravity, and the visuals on this site were made with AI image generation. The physics inside Gravity is not AI: it runs on Rapier, a deterministic rigid-body engine.
- 01
Plan
The owner sets the goal; planning agents turn it into a written plan with success criteria.
- 02
Build in parallel
Implementation agents work in separate branches at the same time.
- 03
Verify in After Effects
Changes are checked inside a real After Effects session, not only in unit tests.
- 04
Keep the evidence
Screenshots, logs and reports from each run are saved next to the work.
How the team is wired
- Aug 12, 2026
The spec is reviewed from six angles and revised 18 times. The first two milestones pass in a real After Effects session.
- Aug 26, 2026
Four parallel worktrees: 3D volume, 2D-to-3D conversion, gizmos, and integration with QA.
- Sep 22, 2026
A full native journey in the installed panel: scan, preview, bake and Undo, with screenshots and receipts.
- Sep 28, 2026
Panel redesign checked with 1,663 automated tests, 147 browser scenarios and native self-tests.
- Oct 4, 2026
1,839 tests and 151 browser QA scenarios pass at one recorded revision.
- Oct 7, 2026
This site: visuals generated with ima2, every claim checked against the product records.
AI agent team
OmO coordinates GPT and Claude agents in separate roles (planning, parallel implementation, research, review and QA), directed by the owner.
Verified in After Effects
Native After Effects runs caught defects that green unit tests missed, and they were fixed before sign-off.
AI-designed visuals
Every illustration on this site was generated with ima2 and is labeled as an illustration.
Gravity contains no AI features. Nothing you simulate is sent to an AI service.
Pricing
Subscribe, or buy it once.
Draft pricing, shown so you know what to expect. Purchasing opens at release.
-
Monthly
For trying Gravity on a project
$12/ month
Draft price
Billed monthly.
-
Yearly
Best valueFor artists who use it every week
$8/ month
Draft price
Billed yearly at $96.
-
Studio
For teams
$384/ year
Draft price
5 seats, billed yearly.
-
Perpetual
For owning a version outright
$199one-time
Draft price
Pay once. Includes 12 months of updates.
Early access
Hear when Gravity opens.
Leave your email and we'll write when the trial and checkout open. We use it only for Gravity launch news.
System requirements
What you need.
- After Effects
- 2025 (25.0) or later. Gravity is a CEP 12 panel, developed and tested in After Effects 2026.
- Operating system
- Windows. macOS has not been verified yet.
- Graphics
- A GPU with WebGL support for the panel preview.
- Interface
- English interface. Korean user guide.
- Output
- Ordinary AE keyframes. Opening a baked project doesn't need Gravity.
The final list of supported versions will be published with the release.
FAQ
Is there AI inside Gravity?
No. The simulation runs on Rapier, a deterministic rigid-body physics engine. AI agents helped us design, build and test Gravity; they don't run inside your After Effects.
Do people I send the project to need Gravity?
No. Output is ordinary AE keyframes.
Are 3D models collided exactly?
They use a single convex hull built from the model's vertices. Concave shapes are approximated; if a hull can't be built, Gravity falls back to a box and says so.
What about text?
Text uses its bounds. Use Split into letters to make each letter of a supported 2D text layer collide on its outline.
Which layers can't be bodies?
Cameras, lights, nulls, audio, guides; plus 3D states like Auto-Orient or per-character 3D, each with a stated reason.
Does it add motion blur?
No. Turn it on in After Effects after baking.
Things pass through each other.
Turn on CCD for fast or thin bodies, or raise the solver rate.
Is anything sent from my machine?
Diagnostics are collected into a local ZIP only when you run it; nothing is sent automatically.
Which After Effects versions?
Development targets After Effects 2025 (25.0) and later on Windows. The final list will be published with the release.
From first drop to final bake.
Physics for your After Effects layers, in 2D and 3D, with keys you can still edit.