# AI plugins for After Effects: choose by the compositing job

Choose After Effects AI tools for roto, removal, restoration, depth and scripting. Compare native features, plugin limits and revision workflows.

Author: Anvisha Pai

Canonical: https://voyageragent.ai/blog/ai-plugins-for-after-effects

Published: 2026-09-15
Updated: 2026-09-17

![A motion designer at a desk at night, one wide monitor showing a layered timeline and a green masked silhouette of a dancer, a stylus in hand.](https://cdn.moda.app/voyager/blog/ai-plugins-for-after-effects/hero-ae-plugins-8dfaa3da28.webp)

Start with Object Matte for subject isolation and Content-Aware Fill for object removal. Add a plugin when the shot needs something more specific: prompt-based selection, editable tracking splines, restoration, depth, or generated footage. For repetitive layer and expression work, use a scripting assistant rather than a video generator.

The distinction is what you can change after the first result. A matte, an editable spline and a generated clip may look similar in the viewer, but leave very different repair jobs. This guide compares documented capabilities and costs; it does not rank roto or enhancement quality from a shared footage test.

**Jump to:** [comparison](#the-comparison-at-a-glance) · [roto](#isolating-a-subject-object-matte-mask-prompter-and-mocha-pro) · [scripting](#scripting-and-agents-the-assistant-two-panels-and-an-mcp-bridge) · [trial checklist](#test-the-repair-not-just-the-first-matte)

## Four layers, and why the layer matters

| Layer                         | What you get back                                                                       | Examples                                                         |
| ----------------------------- | --------------------------------------------------------------------------------------- | ---------------------------------------------------------------- |
| Native tool                   | A matte, fill layer or generated asset inside the project, undoable                     | Object Matte, Content-Aware Fill, AI Assistant (beta)            |
| Plugin in the comp            | An effect or panel result on a layer; the source layer stays                            | Mask Prompter, Mocha Pro, Topaz Video, Depth Generator, Lockdown |
| Panel that places generations | A new clip or image dropped into the comp, generated on a vendor's servers for credits  | Runway plugin, Higgsfield plugin                                 |
| Agent                         | Changes to layers, keyframes, expressions and project structure, made through scripting | AI Assistant (beta), Claude Scripter, AE GPT, Higgsfield MCP     |

An .aep retains layer and effect settings, but footage, generated fill files and installed plugins can remain external dependencies. A parameter may be editable without the project being self-contained. Panels that place generations give you a rendered clip; check whether its prompt and references survive in the saved project, which is the problem covered in [what editable should mean for AI video](/blog/what-editable-means-for-ai-video). Agents are the most powerful and least predictable; the thing to check is whether every action lands in the undo stack.

## The comparison at a glance

| Tool                        | What it does inside After Effects                                                                     | Cost                                                      | The tradeoff                                                                             |
| --------------------------- | ----------------------------------------------------------------------------------------------------- | --------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
| Object Matte (Adobe, 26.2+) | Click or marquee a subject; the matte propagates and can be frozen                                    | Included                                                  | Built for people, animals and vehicles; fine edges still go to Refine Edge or Roto Brush |
| Content-Aware Fill (Adobe)  | Removes masked objects by synthesising pixels from other frames                                       | Included                                                  | Uses surrounding frames; a painted reference can supply missing background               |
| AI Assistant (Adobe, beta)  | Natural-language organisation, expression writing and debugging, Firefly and partner-model generation | Included, beta build only                                 | Not in the release build                                                                 |
| Mask Prompter 3 (Blace)     | Text, point or box prompts produce mattes with Segment Anything 3, locally                            | $49.99                                                    | NVIDIA 8 GB or Apple silicon 16 GB; other Macs unsupported                               |
| Mocha Pro 2026 (Boris FX)   | Planar tracking plus Matte Assist ML, Matte Refine ML, face detection                                 | $37 a month, $295 a year or $695 perpetual (Adobe plugin) | ML tools want an RTX-class GPU; CPU fallback is "significantly degraded"                 |
| Runway plugin               | Generate video and images, restyle a clip with Aleph 2, upscale, remove background                    | Free panel; paid Runway plan and credits                  | Output is a render; check retained generation settings                                   |
| Higgsfield plugin           | Generate video and images, reframe, remove background, Draw to Edit, upscale                          | Free panel; Higgsfield subscription and credits           | Cloud only; plugin generations always spend credits                                      |
| Topaz Video plugin          | Enhance and Frame Interpolation as effects on a layer                                                 | Personal $299 a year, Pro $699 a year                     | Upscaled pixels do not change the comp size; 3x to 4x is slow                            |
| Depth Generator (Red Giant) | Local depth maps from footage for Bokeh, Particular and displacement                                  | Red Giant suite, $85 a month or $639 a year               | Only sold with the suite                                                                 |
| Lockdown 4 (aescripts)      | Mesh tracking of warping surfaces with an AI tracker, 3D export                                       | $399.99                                                   | Price; the free tier stops at two-point tracks                                           |
| Claude Scripter (aescripts) | Prompt to ExtendScript, run and saved to a library                                                    | $39.90 plus your own API key                              | You are running generated code in your project                                           |
| AE GPT (aescripts)          | Chat and Agent Mode that writes and runs ExtendScript                                                 | $29 plus an API key or ChatGPT or Claude subscription     | Same, and Premiere is chat only                                                          |

Prices are from each vendor's page on 15 September 2026, single-user, before tax.

## Isolating a subject: Object Matte, Mask Prompter and Mocha Pro

Rotoscoping is where native After Effects has moved furthest. Object Matte arrived in 26.2 in April 2026, and it is not a brush. You click a subject in the Layer panel, or drag a rectangle or lasso around it, and After Effects generates a matte and propagates it through the shot. Adobe describes it as "built on the same AI technology that powers the Object Mask workflow in Premiere" ([Adobe Help](https://helpx.adobe.com/after-effects/desktop/roto-brush-and-refine-matte/roto-brush/object-matte.html)). The 26.5 release in September added disk caching, so propagated frames survive closing the project instead of living only in RAM ([Adobe, what's new](https://helpx.adobe.com/after-effects/desktop/what-s-new/whats-new.html)).

![Adobe's help page for AI-powered Object Matte, showing the tool menu with Object Matte, Quick Selection, Selection Brush and Refine Edge tools, and a subject outlined in magenta.](https://cdn.moda.app/voyager/blog/ai-plugins-for-after-effects/fig-object-matte-d6cd52cf8b.webp)

Adobe's Object Matte page. The classic Roto Brush is now the Quick Selection Tool in the same menu; Refine Edge is still the tool for hair.

Two things to know before relying on it. It is four tools: Object Matte for the click, Quick Selection (Adobe calls it the "Classic Roto Brush reimagined"), a Selection Brush for additive and subtractive strokes, and Refine Edge for hair. And Adobe's own page sends you back to Roto Brush and Refine Matte "if you need manual control, detailed edge refinement, or a traditional rotoscoping workflow". It is the fast first pass on the subjects it was built for, which CG Channel's coverage summarised as people, animals and vehicles ([CG Channel](https://www.cgchannel.com/2026/04/adobe-releases-after-effects-26-2/)).

Mask Prompter 3 answers a different question: what if the thing you want to isolate is not a person, or you would rather describe it than click it? It runs Segment Anything 1, 2 and 3 plus VITMatte for refinement, and takes text, points or boxes, with multiple objects separated by a full stop in the prompt. The vendor states that "processing happens fully locally" and "your data never leaves your machine" ([aescripts](https://aescripts.com/mask-prompter/)). The purchase check is hardware: the listing's table marks Windows with an NVIDIA GPU of 8 GB or more and Apple silicon on macOS Sequoia with 16 GB of RAM as supported, a non-NVIDIA Windows GPU as "very slow", and any other Mac as unsupported. At $49.99 with a free trial and compatibility from After Effects 2023 to 2026, it is the cheapest prompt-based matting in a comp, if your machine is on the right side of that table.

Mocha Pro is the professional tracking and roto plugin, and 2026 is where its machine learning became a reason to buy. Matte Assist ML generates animated vector mattes, Matte Refine ML produces "clean, natural soft edges from basic matte shapes", and face detection tracks multiple faces for privacy blurs or beauty work ([Boris FX](https://borisfx.com/products/mocha-pro/)). The difference from Object Matte is that Mocha's output is splines and tracking data you can edit, not a raster matte you freeze. The cost is real: $37 a month, $295 a year or $695 perpetual for the Adobe-only plugin, and Boris FX asks for an NVIDIA RTX or AMD Radeon Pro GPU with 8 GB for the ML tools on Windows, noting that otherwise "ML plug-ins can run on the CPU with significantly degraded performance". If your roto is mostly people, try the native tool first. If it is planar surfaces, screen replacements or faces at volume, Mocha earns its price.

## Removing and filling: Content-Aware Fill, and what is not in After Effects

Content-Aware Fill predates the generative wave and works on a different principle. You mask the object and After Effects "analyzes frames over time to synthesize new pixels from other frames" ([Adobe Help](https://helpx.adobe.com/after-effects/using/content-aware-fill.html)). The three fill methods are the thing to learn: Object for moving things like a car on a road, Surface for flat static surfaces like a sign, Edge Blend for textureless areas like text on paper. A Reference Frame round trip to Photoshop lets you paint what the fill should look like when the footage never reveals the background.

If the footage never reveals a usable background, supply a painted reference frame or use another reconstruction method. A generative model is one option, not a requirement. Adobe's IBC announcement on 8 September put Generative Extend and the new Generative Media tool, with Firefly plus partner models from Google, Kling, Runway and Luma, in Premiere Pro, and gave After Effects the AI Assistant beta instead ([Adobe blog](https://blog.adobe.com/en/publish/2026/09/08/generate-create-directly-in-your-timeline-with-new-ai-powered-innovations-in-premiere-after-effects)). Inside a comp, the generative routes to fill or extend are third-party: Higgsfield's Draw to Edit, which sketches a region to remove or add elements, and Runway's Aleph 2, which re-renders a clip from a restyled anchor frame.

## Generating footage into a comp: Runway, Higgsfield and Adobe's Assistant

All three options here share a shape: a panel inside After Effects, generation on a server, credits from an account you already have, and a rendered result placed in the project.

Runway's plugin shipped on 8 September 2026. It opens from Window, Extensions, Runway, and offers Runway's video models (Gen-4.5, Seedance 2.5, Kling 3.0 Pro, Veo 3.1), image models (Gen-4 Image, Nano Banana, GPT Image 2), upscaling, background removal and HDR conversion ([Runway plugins](https://runway.com/plugins)). The feature specific to a comp is Edit Studio: select a clip, restyle its anchor frames, and "Aleph 2 re-renders the full clip to match the same duration as the source" ([Runway](https://runway.com/news/company-news/runway-for-adobe)). The panel is free; generating uses the credits on the plan you already have, and the changelog entry is labelled for paid plans ([Runway changelog](https://runway.com/changelog)).

![Runway's plugins page: 'Generate, edit and extend footage with Runway's models directly inside Premiere Pro and After Effects. No exports, no round trips.'](https://cdn.moda.app/voyager/blog/ai-plugins-for-after-effects/fig-runway-plugins-34b1adf1b9.webp)

Runway's plugin page. The claim is about the round trip, not the model: the models are the same ones on the website, billed from the same credits.

Higgsfield's plugin has been available since July 2026 and covers Generate Video, Generate Image, Reframe, Remove Background, Draw to Edit, Upscale and Edit Video ([Higgsfield](https://higgsfield.ai/blog/higgsfield-after-effects)). Two conditions from the help centre matter for budgeting: it works on After Effects 2024 (24.0) and newer, and while it shares your web credits, "unlimited access and free generations apply only on higgsfield.ai", so every plugin generation spends credits even on a plan that is unlimited in the browser ([Higgsfield help](https://higgsfield.ai/creator-hub/help-center/integrations/external-integrations-higgsfield)). It also needs a connection and sign-in, because inference runs on Higgsfield's servers.

Adobe's AI Assistant is the third route, but only in the beta build. Its help page lists generating "footage, reference images, or design elements with Adobe Firefly, directly in the Assistant panel", and video "from reference images or text descriptions with Adobe Firefly and partner models" ([Adobe Help](https://helpx.adobe.com/after-effects/desktop/work-with-after-effects-ai-assistant/after-effects-ai-assistant-overview.html)).

Whichever panel you use, the generation is a clip. Whether to start it from a still or a text prompt is covered in [text-to-video vs image-to-video](/blog/text-to-video-vs-image-to-video); for compositing, start from a frame of your own comp whenever identity, framing or continuity matters. Keep the prompt, reference and model with the project; verify what your chosen panel preserves when you reopen it.

## Upscaling and restoration: Topaz and the generation panels

Topaz Video is the specialist, and its plugin is narrower than the app. The docs say the After Effects Effect Manager should show "Enhance and Frame Interpolation", and the plugin is activated through the main Topaz Video application ([Topaz docs](https://docs.topazlabs.com/topaz-video/plugins)). The part that trips people up is stated plainly: when upscaling 2x or more, "the project or timeline will not take effect", so the layer is oversized until you scale it or build a larger comp, and "performance and UI responsiveness may be reduced when upscaling to 3-4x".

![Topaz Video documentation for the After Effects plugin, showing the Plugins menu in the Topaz app and a list of third-party application considerations, including that the project or timeline size will not change when upscaling.](https://cdn.moda.app/voyager/blog/ai-plugins-for-after-effects/fig-topaz-plugin-docs-4aba4db16b.webp)

Topaz's plugin page. The install goes through the Topaz app, and the caveats about comp size and 3x to 4x performance are in the vendor's own words.

Pricing is subscription only: Personal is $299 a year with unlimited local rendering, 25 cloud credits a month and commercial use limited to organisations under $1M in annual revenue; Pro is $699 a year with more models, 100 cloud credits and full commercial rights ([Topaz](https://www.topazlabs.com/topaz-video)). For occasional upscales, the Runway and Higgsfield panels include an upscale tool on credits you already pay for; for restoring an archive, Topaz's unlimited local rendering is the cheaper shape. If you already own Boris FX Continuum, 2026.5 added BCC+ DeInterlace ML and updated UpRes ML and Motion Blur ML models ([Boris FX](https://borisfx.com/news/continuum-2026-5-ai-precision-more-creative-control/)), which may cover the restoration you were about to buy separately.

## Depth and tracking: Depth Generator and Lockdown 4

Red Giant's Depth Generator, added in Red Giant 2026.1, "analyzes footage or images and produces a grayscale representation of the depth of the shot for use with other effects", and runs "entirely on your local machine" ([Maxon](https://www.maxon.net/en/product-detail/red-giant/vfx/depth-generator)). Maxon's knowledge base names the model as Depth Anything V2, installed as a self-contained file that "does not rely on external servers or cloud-based processing" ([Maxon support](https://support.maxon.net/hc/en-us/articles/22124050051228-Depth-Generator-AI-Information)). The pass feeds Bokeh for rack focus and Particular's ZBuffer so particles pass behind a subject. The catch is licensing: it only comes with the Red Giant suite at $85 a month or $639 a year, a good deal if you use Trapcode and a poor one if you only want depth.

Lockdown is the mesh tracker for warping surfaces, and version 4 is where AI entered its pipeline. The listing describes "an ai based tracker that is much smarter than our previous efforts" and a search-and-fix feature that snaps a drifting point to the right feature using other frames as reference, plus 3D export to Cinema 4D and, through Alembic, to Blender and Maya ([aescripts](https://aescripts.com/lockdown/)). It is $399.99, and the free version stops at one- and two-point null tracks. It does not compete with Object Matte; it is what you reach for when a logo has to stay stuck to a moving shirt.

## Scripting and agents: the Assistant, two panels and an MCP bridge

Here "AI plugin" means a model that writes ExtendScript and expressions and runs them against your project. Four options differ mostly in who is accountable for the code.

Adobe's AI Assistant, in the beta, does three things according to its help page: organise projects (rename, restructure, colour-code), work with expressions (write from a description, explain one in plain language, diagnose an error), and generate media ([Adobe Help](https://helpx.adobe.com/after-effects/desktop/work-with-after-effects-ai-assistant/use-after-effects-ai-assistant.html)). Two defaults are the right ones: it "asks for your permission before it takes an action in your project", with a setting to approve automatically, and undo "steps through one change at a time, rather than the whole request at once". It is the only agent here that Adobe supports, and the only one not yet in the release build.

Claude Scripter is a $39.90 panel that turns a sentence into ExtendScript, runs it, and saves it to a library. Version 1.3.0 from July 2026 supports Claude, OpenAI, Gemini and Grok models plus Ollama and LM Studio for local ones, and works back to After Effects CC 2018 ([aescripts](https://aescripts.com/claude-scripter/)). You bring your own API key, and the vendor states that every generated script is wrapped in an undo group so one Ctrl or Cmd+Z reverts it ([Claude Scripter](https://claudescripter.com/)). Its listed sweet spot is housekeeping: batch renames, unused solids, layout, loopOut on a selection.

AE GPT is $29 and goes further with Agent Mode, which "writes and runs ExtendScript inside your compositions automatically" after reading the active comp's layers, effects and expressions. It supports OpenAI, Anthropic and Google models and local Ollama, and since 1.2.1 you can sign in with a ChatGPT Plus or Pro or a Claude Pro or Max subscription instead of an API key ([aescripts](https://aescripts.com/ae-gpt/)). The vendor is clear about scope: it "does not cut footage, edit audio tracks, mix music, or replace a full NLE workflow", and Premiere support is chat only ([AE GPT](https://ae-gpt.com/)).

Higgsfield's MCP is for people who already work in Claude, Cursor or a coding agent. Install the Adobe plugin, add the connector at bridge.higgsfield.ai/mcp, and the agent can build compositions, set keyframes and write expressions in the open project ([Higgsfield](https://higgsfield.ai/plugins/after-effects)). There is also an MIT-licensed local bridge on GitHub that talks to After Effects directly, needs Node 24 and the scripting file-access preference, and does not require a Higgsfield account. Its README carries the caveat every agent user should read: no automatic rollback on a failed operation, and a timeout can mean a mutation ran without confirmation ([GitHub](https://github.com/higgsfield-ai/fnf-local-pluging-bridge-mcp)). Every server and scripting route for After Effects is compared in [After Effects MCP](/blog/after-effects-mcp).

The purchase check for all four is the same. Generated ExtendScript runs with the same authority as a script you wrote. Prefer the tool whose actions land in the undo stack in a way you understand, save incrementally, and keep the agent's job narrow: rename, organise, wire an expression, build a rig you then inspect. Describing that job well is the same discipline as [briefing an AI video agent](/blog/how-to-brief-an-ai-video-agent): the outcome, the constraints, and what done looks like.

## What a scripting revision leaves editable

We ran a small native workflow on After Effects 26.5 (build 89) on 17 September 2026. Using Voyager's local scripting integration, we built two three-second compositions from text and solids, changed only the end-card headline, and saved a separate .aep. Frames rendered at 0.75 and 4.25 seconds showed the unchanged first shot and revised title; the first pair of PNGs was byte-identical.

![After Effects with our six-second test project open, two composition layers in the timeline and the revised JOIN THE WAITLIST headline.](https://cdn.moda.app/voyager/blog/what-editable-means-for-ai-video/ae-editable-timeline-890fcf183b.webp)

Our local After Effects 26.5 test. The change targets a text layer; it does not regenerate footage. This is a scripting-workflow check, not a comparison of the plugins listed above.

That is the useful difference between a script and a generated clip: the script can leave a headline as text you can address again. The [original and revised projects](https://cdn.moda.app/voyager/blog/what-editable-means-for-ai-video/editability-projects-6e861c2032.zip) contain no external footage or third-party effects. The result does not establish roto quality, plugin reliability or animation-wide consistency.

## Test the repair, not just the first matte

Choose a short shot with an occlusion, a fast movement and a difficult edge. Run the native tool and the plugin trial on the same range. Put the matte over a contrasting solid and inspect the bad frames at full resolution. Record how many corrections are needed and whether you can fix a single frame without restarting propagation.

For restoration, judge the delivered comp rather than an enlarged preview: check edge halos, texture that flickers between frames, and render time at the target size. For a scripting assistant, start with three named layers, make one requested change, and check Undo. These are suggested acceptance tests, not measured results for the plugins above.

## How to choose

Start with what you already pay for. On 26.2 or later, Object Matte and Content-Aware Fill cover most roto and removal without another purchase, and the beta build adds an Assistant that writes expressions and generates Firefly media. Add Mask Prompter to isolate things that are not people, if your GPU is on its list. Add Mocha Pro when the job is planar tracking, screen replacement or editable spline mattes at volume.

For generated footage, choose by the credits you already have: Runway's panel for its models and Aleph 2 restyling, Higgsfield's for reframe and Draw to Edit, remembering that plugin generations always cost credits. Both place generated footage in your comp; distinguish revising that layer from regenerating the shot. Keep the prompt with the project. For a large restoration job, compare Topaz's local-rendering subscription with the cloud credits required for the actual footage. Depth Generator makes more sense if you already use Red Giant. For scripting, compare Claude Scripter and AE GPT on model costs, undo behavior and the tasks their APIs can reach; Adobe's beta availability is a separate constraint.

## Frequently asked questions

**Does After Effects have built-in AI, or do I need plugins?** More than most people assume. Object Matte (26.2 and later) does one-click subject isolation, Content-Aware Fill removes masked objects by borrowing from other frames, and the beta build has an Assistant that writes expressions, organises projects and generates media with Firefly and partner models. Plugins are for what those do not cover: text-prompted matting of arbitrary objects, planar tracking, upscaling, depth maps, mesh tracking, and generation from a specific vendor's models.

**Is Generative Extend available in After Effects?** Not as of September 2026. Adobe's IBC announcement put Generative Extend and the Generative Media tool in Premiere Pro. In After Effects, the closest routes are the Assistant beta for generation, Runway's Aleph 2 restyle through its plugin, and Higgsfield's Draw to Edit, all of which produce a rendered clip rather than extending a layer in place.

**Do AI plugins for After Effects run locally or in the cloud?** Both, and the vendor says which. Mask Prompter, Depth Generator, Topaz's local rendering and Mocha Pro's ML tools run on your machine, which is why they publish GPU requirements: Mask Prompter wants an NVIDIA GPU with 8 GB or Apple silicon with 16 GB of RAM, and Boris FX asks for an RTX-class card for Mocha's ML tools. Runway's and Higgsfield's panels and Adobe's Firefly generation run on servers and bill credits. Local tools cost more up front and nothing per use; cloud tools cost nothing to install and something per generation. On an Intel Mac, check each listing before buying; several exclude it.

**Can an AI agent write After Effects expressions and scripts for me?** Yes, through Adobe's Assistant in the beta build, Claude Scripter or AE GPT from aescripts, or Higgsfield's MCP connector from an agent like Claude. All generate ExtendScript or expressions and run them in your project. Check how each handles undo: Adobe's steps through one change at a time and asks permission by default, Claude Scripter wraps each script in an undo group, and the open-source Higgsfield bridge documents that it has no automatic rollback.

**Is there an AI motion capture plugin that works inside After Effects?** No vendor documents one that runs in the comp. Markerless capture tools produce skeletal data for a 3D application, and the round trip goes through Cinema 4D, Blender or a rendered sequence. Inside the comp, the AI tools that touch motion are trackers: Mocha Pro's Point Track ML and Lockdown 4's mesh tracker.


## Make your next project in Voyager

Create an account, download Voyager, and start making.

[Sign up](/sign-up)