October 2, 2026 · 5 min read
Atmosmotion v0.4: motion graphics built by AI agents, for ads
Atmosmotion v0.4 lets an AI agent write a motion scene as code, preview it and render video ads and reels. What shipped, render speed and how we measured it.
By the AdvisorPPC Team · Reviewed by Claude

Atmosmotion v0.4 is out. Atmosmotion is the AdvisorPPC motion engine for AI agents: your agent connects over MCP, writes a motion scene as code, checks a preview and renders video ads, motion graphics and reels. The headline of this release is code scenes, and the rest of this post is what they are, how fast they render, and how we measured the quality.
What is new in Atmosmotion v0.4?
Code scenes. Until now an agent filled templates and presets, which capped the result at whatever the template author imagined. In v0.4 the agent writes the animation itself: a draw function on the Canvas 2D API or WebGL2, called once for every frame, against a motion kit built for ad work.
- Easing, analytic springs, stagger and timelines for timing that feels designed.
- Shape morphs, 3D projection, 3D text rings and glyph particles for type that moves like a material.
- Seeded noise, OKLab color mixing and halftone for texture without stock footage.
- Brand tokens, params and variants, so one scene renders at 16:9, 9:16 or 1:1 with your colors and copy.
- Real temporal motion blur, and an audio track you supply muxed into the render.
How does an agent build an ad with it?
Four tools, in order. code_scene_guide hands the agent the contract, the kit reference, craft rules and a worked example. code_scene_validate catches flat frames, scenes where nothing moves, missing fonts and blocked requests. code_scene_preview returns a contact sheet as an image, with a timecode on each frame, so an agent that can see images checks its own work before rendering. code_scene_render writes MP4, WebM, MOV, GIF, PNG or a PNG sequence.
Renders are deterministic. The engine owns the clock and replaces randomness inside the render, so the same scene produces the same pixels every time, and parallel workers match a single worker byte for byte. Each scene runs sandboxed with network requests blocked apart from web fonts, and a scene that loops forever fails with a clear error instead of hanging.
How fast does it render?
Our benchmark scene is 15 seconds at 1920 by 1080 and 30 frames per second, with 8-sample motion blur. In v0.4 it renders in about 26 seconds on 4 cores. The same scene on a single worker moving raw frames took 229 seconds. The gain comes from ordered parallel workers and lighter frame transport. Your own scene will take more or less depending on its length, size and what it draws.
| Render path | Time |
|---|---|
| Single worker, raw frames | 229 s |
| Ordered parallel workers, JPEG frames, 4 cores (v0.4) | about 26 s |
Is the motion any good?
We tested it blind. A fresh agent got only the Atmosmotion guide and the tools, was told not to read any existing scene, and was asked for a 15-second showreel. It ran 2 validations, 5 previews and 1 render. A separate agent then scored four unlabeled clips on a 100-point motion rubric covering hook, idea, hierarchy, type, color, motion, structure, craft, accessibility and delivery.
| Clip | Score |
|---|---|
| Scene written by the fresh agent with Atmosmotion | 75.0 |
| Atmosmotion example reel | 72.5 |
| External reference clip | 72.0 |
| Earlier hand-written HTML reel, no engine | 63.5 |
Two honest limits. Both the author and the judge were AI models from the same family, and a review by an independent judge is still pending. And no clip reached the rubric's 80-point ready line. The judge's notes have already changed the kit: an overlay pass so captions no longer ghost under motion blur, plus velocity, squash-and-stretch and impact helpers. Stepped blur on very fast moves is still open.
Do I need After Effects?
No. Atmosmotion renders on its own engine, so no After Effects license is needed. What you need is an AI client that can add a remote MCP server.
Where does this fit in an ad workflow?
Every placement wants its own cut. Atmosmotion makes the cuts; the campaign work stays where it was. If you are planning video, Video action campaigns and Demand Gen campaigns are Google formats that run on video, and the words on screen still follow the rules in how to write Google Ads that get clicks. Nothing your agent renders goes live by itself: publishing an ad is a separate step, and approval gates are how AdvisorPPC keeps that step in your hands.
How do I try it?
Open the Atmosmotion connector page, add Atmosmotion to your AI client as a remote MCP server, sign in when the browser opens, and brief your agent. The films and the FAQ are on the Atmosmotion page.
Benchmark and review figures come from the Atmosmotion release notes and the blind review write-up dated 2026-09-30. The review used AI author and judge agents; an independent judge is pending. Atmosmotion has no separate price list yet; AdvisorPPC plans are on the pricing page.