AI for Motion Designers in 2026: Where It Fits in an After Effects Workflow

In 2026, AI fits into an After Effects workflow in three places: language models that write expressions and scripts, code-first tools such as Remotion and HyperFrames that render motion an AI agent writes as code, and generative models that supply raw material such as backgrounds, textures, b-roll, loops and cut-outs. It does not replace the comp. Kerning, frame-exact typography timing and brand-exact vectors still belong to you and to After Effects.

Updated October 2026.

Disclosure: VdoBloom is our platform.

The short version: AI feeds the comp, it doesn’t replace it

Motion work splits into making assets (plates, textures, footage, mattes) and choreography (timing, easing, hierarchy, type). AI is now good at the first and a useful code-writing assistant on the second. It is weak wherever the result must match a spec to the pixel or the frame.

LLMs writing expressions and ExtendScript

This is the cheapest win. ChatGPT and Claude write working After Effects expressions if you tell them the property, the expression engine, and what the value should do over time.

The engine matters. Since version 16.0, After Effects has a JavaScript engine (ECMAScript 2018) and a Legacy ExtendScript engine (ECMAScript 3), set in File > Project Settings > Expressions. Some syntax works in only one: Math.random() is missing from Legacy, while random() works in both. Naming the engine avoids most broken answers.

A good request looks like this: “Position property, JavaScript engine. A wiggle that loops seamlessly every 4 seconds so I can render a looping background.” A capable model returns the standard cross-fade pattern:

freq = 1; amp = 40; loopTime = 4;
t = time % loopTime;
w1 = wiggle(freq, amp, 1, 0.5, t);
w2 = wiggle(freq, amp, 1, 0.5, t - loopTime);
linear(t, 0, loopTime, w1, w2)

ExtendScript (.jsx) scripts pay off even more: renaming hundreds of layers, building lower thirds from a CSV, batch-queuing renders. Run generated scripts on a copy of the project, because a model can invent an API call.

Adobe is building this in too. In September 2026 it put an AI Assistant into After Effects as a public beta. Adobe says it works across a whole project, reorganises large layer stacks and can create effects such as expressions without code. Treat it like any beta.

Code-first motion: Remotion, HyperFrames and AI agents

The second shift is motion graphics written as code, which coding agents are good at producing. Two open projects matter here.

Remotion builds videos as React components and renders them to video files. Its license is not a standard open-source license. It is free for individuals, for for-profit companies with up to 3 employees, for non-profits, and for evaluation. Larger for-profit companies need a company license.
HyperFrames, from HeyGen, is Apache 2.0 licensed. You write a video as HTML with data attributes for timing, animate it with GSAP, CSS, Lottie or Three.js, and render it to MP4 by stepping a headless Chrome through each frame and encoding with FFmpeg. The project ships agent skills, so a coding agent can plan, write, preview and render a piece on its own.

Code-first motion suits data-driven or templated work: a chart animated from a spreadsheet, 40 localised versions of one promo. The agent writes the template once and every render is deterministic. Keep one-off hero pieces, where hand-tuned easing matters most, in After Effects.

MCP servers: letting the agent fetch footage too

Model Context Protocol (MCP) lets an assistant such as Claude, ChatGPT or Cursor call outside tools, so the agent that writes your template can also produce its assets.

VdoBloom runs an MCP server at https://vdobloom.com/mcp. You add it as a custom connector and sign in through the browser; there is no key to paste. Command-line clients can use a vdb_ key instead. It exposes seven tools: list_models, get_balance, generate_image, edit_image, generate_video, get_status and list_creations. An image on the default Nano Banana model costs 6 credits. A 5-second Seedance 2 Fast clip costs 24 credits at 480p and 50 at 720p. The assistant can read prices before it spends anything. Setup is covered in how to generate AI images and videos from Claude or ChatGPT, and the developer page documents the REST API as well.

Example: ask for three gradient plates at 16:9, animate the best into a 5-second drift, and drop it into the composition the agent just wrote. Effects, lip-sync and timeline editing are not exposed; those stay in the app.

Where generative models genuinely help

Generated pixels earn their place as material you composite, not as the finished piece:

Job	VdoBloom tool	Credits	Handoff to After Effects
Abstract backgrounds, textures, style frames	AI image generator (Seedream 5 Flash)	2 per image	Import as a plate; grade in AE
Elements with real transparency	Qwen Image 2.1	3 at 1K, 5 at 2K	Transparent PNG layer
Parallax from a single still	Split image into layers	16 per run	One PNG per layer; offset in 3D space
B-roll and animated plates	Image to video or text to video	from 24 (5 s, 480p, Seedance 2 Fast)	MP4 footage layer
Seamless loops	Seamless loop video	28 from a photo (Kling 3.0, 5 s, 720p); 10 to blend an existing clip	Loop the footage layer
Logo sting draft	AI logo animation (Veo 3.1)	160 per ~8 s clip	Reference or rough cut; rebuild the final in AE
Cut out a still	Background remover	1 (standard), 2 (HD)	PNG with alpha
Cut out footage	Video background remover	4 per second, up to 30 s	WebM with alpha, or green-screen MP4 to key
Upscale stills or footage	Image upscaler / video upscaler	from 4 per image; video 3.2/s at 1x–2x, 5.6/s at 4x, 30 minimum	Replace the source layer

If your After Effects version won’t import the transparent WebM, choose the green-screen MP4 and key it with Keylight. The logo tool accepts raster files only (PNG, JPG or WebP, up to 10 MB), not SVG, which is another reason to treat it as a draft. See our seamless loop guide, upscaling guide and b-roll guide for detail.

Where AI still doesn’t fit

Typography timing. A video model cannot land a word on frame 37 and ease it out over 8 frames, and generated text still warps. Set type in AE.
Brand-exact vectors. Generators output raster pixels and redraw a logo rather than move your paths. An AI sting is a mood piece, not the approved asset.
Exact repeatability. The same prompt twice gives two different clips. Same move, new headline is an expression or template job.
Camera data. Generated footage has no reliable camera solve, so you track it yourself.

Best AI tools for motion designers

Facts below were checked on each vendor’s own site in October 2026 unless noted. Prices change often, so confirm on the vendor’s pricing page.

Tool	Best at	Pricing model
Adobe After Effects AI Assistant	Project clean-up and expressions inside AE (public beta)	Part of After Effects; check Adobe for plan details
Runway	Generation and editing in one web app; Gen-4.5, Aleph and partner models	Free tier with 125 one-time credits; Standard $15/month ($12 yearly)
Kaiber	Canvas-based Superstudio and audio-reactive visuals for music	Check Kaiber’s pricing page
Cavalry	Procedural 2D motion and Lottie export (not generative AI)	Free to download and use; now backed by Canva
Rive	Interactive animation with state machines for apps and web; includes an AI agent	Free tier; paid seats from $9/month
Remotion / HyperFrames	Motion as code, written by coding agents	Remotion: free for small teams, company license above; HyperFrames: Apache 2.0
VdoBloom	Many image and video models plus utility tools (loops, keys, upscale, layers) in one place, with an MCP server	Pay-as-you-go credits that never expire; no subscription required

For a deeper look at two of these, see Higgsfield vs Runway. If you are choosing a still-image model for style frames, see best AI image generators in 2026.

Verdict by use case

Freelancer doing explainers and promos: use an LLM for expressions every day, and generate plates and textures for a couple of credits each. Keep type and logos in AE.
Studio producing templated or data-driven work: move the repeatable parts to Remotion or HyperFrames and let a coding agent maintain them. Check Remotion’s license against your headcount.
Designer building an agent pipeline: connect an MCP server so the same agent that writes the code also fetches the assets.
Music visuals and VJ loops: Kaiber’s audio-reactive tools, or VdoBloom’s seamless loop tool for clips that repeat cleanly.
Interactive product motion: Rive, with AE kept for marketing renders.

FAQ

Can ChatGPT or Claude write After Effects expressions?

Yes, reliably for common patterns, if you name the property and the expression engine (JavaScript or Legacy ExtendScript). Test scripts that change a project on a copy first.

Is Remotion open source?

Its source is public, but the license isn’t a standard open-source one. It is free for individuals, companies with up to 3 employees and non-profits. Larger companies need a company license. HyperFrames uses Apache 2.0.

Can AI animate my logo exactly as designed?

Not exactly. Video models redraw the mark from a raster image. Use an AI sting to explore ideas, then build the final from your vectors in After Effects.

Do I need a subscription to use VdoBloom for motion assets?

No. Credits are pay-as-you-go and never expire, and new accounts start with 10 free credits, enough to test a few plates before deciding.

To try it, generate one texture in the AI image generator, turn it into a drifting plate with image to video, and drop it into your next comp. Check the pricing page for current credit packs.