Adobe has opened Creative Cloud to third-party AI agents. Connectors for Claude, ChatGPT, and Gemini already work, and at Adobe MAX the company previewed local MCP integrations that will let those agents edit native Premiere and After Effects project files on your machine.
Two other tools approach the same job from different directions. Remotion lets a coding agent such as Claude Code write a video as React code, and CoAnimator lets an agent build and edit a video inside a desktop studio. The job is familiar to anyone who ships software. A product video starts going stale the day the product ships, and someone has to rebuild it.
Where the agent works decides who on your team can use it, what a revision costs, and whether it works today. Each of the three comes with a catch, and the catches are different. I have analysed all three to see where each works best, where its limits are, and which one is the strongest agentic AI choice for each kind of video.
Key Terms for AI Video Workflows
Four terms first, so the rest reads without a search tab.
- Agent: A coding or chat assistant such as Claude Code, Codex, or ChatGPT that can read files, run commands, and call outside tools.
- MCP: The Model Context Protocol, a standard that lets an agent use another application as a tool.
- Agent skill: A set of instructions that teaches an agent how to work inside a specific tool.
- Project file: The file that holds your video, such as a Premiere sequence, a React project, or a CoAnimator project.
Adobe lets the agent work in the native files of its pro apps, Remotion in a code project, and CoAnimator in a studio built for agents.
Adobe MCP for Premiere and After Effects
Adobe's local MCP integrations are a preview, announced at Adobe MAX, that will let the third-party agent you already use work in Photoshop, Illustrator, Premiere, and After Effects. The agent changes native project files directly on your computer. Layers, vectors, bins, and sequences stay editable, and you can step in at any moment.
Adobe already offers a different kind of integration today. Adobe's connector for Claude lets Claude use more than 50 tools from Photoshop, Illustrator, and Premiere inside Anthropic's cloud-based service, and similar integrations exist for ChatGPT and Gemini. You drop a file into a conversation, the agent works on it, and you continue in Adobe's apps later. The local MCP reverses that. The agent comes to your files.
- Agents it names: Coverage of the announcement names ChatGPT, Claude, and Gemini as agents that could drive the apps.
- Native files: The agent edits the real project, so the work stays in Adobe's own format.
- Connected workflows: If the agent is also connected to other MCP tools such as Google Drive or Slack, it can pull a brief from Google Docs, work in Adobe's apps, and share the result in Slack.
- Adobe's own agent: Adobe Creative Agent is available in Premiere, Illustrator, InDesign, After Effects, and Frame.io, with AI assistants in beta. In Premiere it can apply effects across every clip in a sequence, track masked subjects through a frame, and adjust color and exposure across a whole sequence.
- Generation in the timeline: Premiere's Generative Media Tool generates video, sound effects, soundscapes, and music directly in the timeline.
- You stay in control: Adobe says you choose when to delegate work to the agent and when to make hands-on adjustments yourself.
- Availability: The local MCP integrations arrive in private beta in early 2027.
Adobe's route is for editors who already live in its apps and want an agent to take the repetitive work. I would wait for the beta notes before building a workflow on it, since Adobe has not yet said how liability, provenance, or cost will work. Remotion approaches the same question from code.
Remotion for AI-Powered Video Creation with React
Remotion is a framework for making videos with React. Its Agent Skills teach coding agents such as Claude Code, Codex, Kimi Code, and OpenCode how to work in a Remotion project.
Remotion's own launch post for those skills shows 18.8M views on X.
Remotion maintains an official plugin for Claude Code. In the terminal, you install it with two commands.
claude plugin marketplace add remotion-dev/claude-code-plugin
claude plugin install remotion@remotion
After you restart Claude Code and open a session in a new folder, the plugin's create skill takes your prompt.
/remotion:remotion-create Make a promo video for a record store
- Code is the source of truth: Remotion describes three ways to work, agentically with your coding agent, interactively with drag and drop, and programmatically with data and code, and you can switch between them.
- Remotion Studio: A fully interactive editor where you select, drag, trim, crop, and edit, with Fast Refresh so changes show up as you save.
- Agent skills and plugin: Official skills cover creating a project, previewing in Studio, rendering, captions, maps, and multimedia, and WebMCP lets an agent control Studio. Adding .md to any doc URL returns raw markdown for agents.
- Prompt to Video: A template that turns prompts into short videos with a script, images, and voiceover.
- Player: Embed a video written in React inside your web app and change it at runtime from data or a form.
- Rendering: Render from the CLI, Node.js, or the cloud. Remotion Lambda renders pieces of a video in parallel on AWS, Google Cloud Run is supported, and outputs include MP4, WebM, GIF, and stills.
- Starters and packages: Templates for Next.js, React Router, Remix, and Express, plus packages for 3D, shapes, Tailwind, and Rive.
Remotion's strength is control. Because the video is code, every frame can come from data and logic, and the same project can power a Player inside your product. The price is that you maintain a React project, even when the agent writes it. I see that trade as fair for a team that already lives in React, and expensive for a team that does not.
The next tool also keeps the video editable, but as a studio project instead of code.
CoAnimator for Agentic Video Editing
CoAnimator is an agentic animation studio where an AI agent builds and edits videos from your description, such as a product demo, launch video, tutorial, or explainer. The agent you already use researches the product, drafts the script, and builds it scene by scene while you watch and redirect.
The above video, "Life Leaves a Trace," comes from CoAnimator's own YouTube channel. It is an animation built from a storyboard, animated scenes, narration, music, and sound effects inside one editable project. Here is what the studio looks like while an agent works.
CoAnimator does not ship its own chat model. The intelligence comes from the agent you connect, and CoAnimator supplies the studio around it.
- Describe, don't code: You describe the video, and the agent writes the animation, timeline, and narration.
- Bring your own agent: Claude Code, Codex, Antigravity, OpenClaw, Gemini, or any agent that can read files and run shell commands.
- Live stage, editable files: The preview updates as the agent works, and the video lives as git-friendly project files you can change and re-render later.
- Templates and skills: Starting points for each video type, each with a skill that teaches the agent the format.
- Kinds of video: Demos, launches, tutorials, explainers, motion graphics, low-poly 3D, and effects over real footage, including AI-generated footage. Scripts, captions, and narration can be in other languages, depending on your voice provider.
- Research through MCP: MCP tools give the agent keyword research, Google Trends data, SEO checks, and web scraping before it writes the script. Research mode is part of the lifetime license.
- Plugins and platforms: Voice, music, screen capture, and publishing, including a Postiz plugin. Standard plugins work on the free plan with some limits, and premium plugins come with the license. It runs on macOS, Windows, and Linux with no account.
CoAnimator's own guidance is to ask for a scene plan and sample frames before the full build, which makes a wrong scene easier to catch while the agent works. I would follow that advice on a first project.
Next, the evidence behind each tool.
Evidence of AI Video Workflows in Action
Here is what each one supports and what it does not, so you can weigh it yourself.
Adobe's announcement
Adobe's own release describes the preview and a private beta, and it publishes no customer results for the local MCP. Independent coverage is cautious. Fast Company asks what happens if a third-party agent corrupts a sequence, generates something with licensing problems inside a project, or quietly burns through generative credits. Asked who is liable, Adobe pointed back to the preview status, and it said it is still working through how provenance will be supported and exploring upfront cost estimates.
Remotion's published results
Remotion's evidence is the deepest of the three.
- Launch reach: Remotion's own post announcing Agent Skills shows 18.8M views on X, and Remotion published the full prompt history behind the animation.
- Public projects: Remotion's resources page lists real projects, including GitHub Unwrapped, a personalized year-in-review, Podcast Maker, and Product Hunt Today, and the repository points to Fireship's "This video was made with code."
- A candid user account: One developer who tried it after the launch reported that a product launch video took multiple days, that vibe coding in Remotion felt 10 to 20 times slower than building app code, and that the first results looked bland and did not improve over ten prompts.
The launch numbers measure reach, and the showcase projects are Remotion's own picks. The developer's account shows how much iteration good results can take.
CoAnimator's published results
CoAnimator's demo videos come with build times.
- Timings: Its effect demos report about 5 minutes from one prompt for a freeze-and-pop effect, against 1 to 2 hours by hand, and about 3 minutes for a cinematic grade, against 30 to 60 minutes.
- Real use: A launch video for a Copilot SDK release, which the page lists at 417K views on X.
- Full demos: A narrated product demo is on YouTube, and a motion-graphics explainer for a chatbot widget is on the page.
These timings are based on my own use, and the by-hand comparison is an estimate. The view count measures reach, not independent quality.
CoAnimator Community Projects and User Feedback
Beyond CoAnimator's own pages, public posts and curated lists show people using it.
- One user: He posts CoAnimator animations on X and often shares the project files. A curated GitHub list of Opus 5.5 demos includes his coded space film "A Cosmic Journey," which he built in CoAnimator from a narration clip over five or six iterations in about 20 to 30 minutes. A second list credits him with a launch video built with Opus 5.5 agents inside CoAnimator within an hour.
- Another user: Posted a 2-minute Black Friday film with AI-made shots, voice, and sound effects, edited entirely in CoAnimator.
Two of user's posts show the range, a 3D keyboard scene he offered the files for and a motion-graphics demo he framed as an After Effects replacement.
One more post comes from another user, the Black Friday film.
How the Evidence Compares
Adobe has the biggest name and nothing you can test yet. Remotion has the most established proof, including a launch post with 18.8M views, a public showcase, official agent skills, and a candid user account that shows how much iteration good results take. CoAnimator is a young product. Its proof is its own demos and timings, plus a small set of community posts, with Rege the most visible user. That shows range and early adoption, and independent head-to-head testing.
The table below puts the three side by side.
Adobe MCP vs Remotion vs CoAnimator Compared
Each row is a difference that tends to decide the choice, so look for the rows that match your own work.
| Question | Adobe MCP | Remotion | CoAnimator |
|---|---|---|---|
| What the agent works on | Native Premiere, After Effects, Photoshop, and Illustrator files | A React code project | A CoAnimator project, kept as files |
| Available today | Cloud connectors now, local MCP in private beta in early 2027 | Yes | Yes |
| Best for | Pro editors who already work in Adobe's apps | Developers putting video inside a product or rendering from data at volume | Product demos, launches, tutorials, explainers, and effects over footage |
| Skill needed | Adobe app knowledge, with the option to step in by hand | React and a terminal, even when the agent writes the code | No coding. You describe the video |
| Research and script | Whatever MCP tools your agent has connected | Agent skills and a Prompt to Video template, with no research step described | Agent researches and drafts, with research mode for keywords, trends, and SEO checks |
| Revising | Edit the native file by hand or ask the agent | Change the code, then re-render | Edit a scene, then re-render |
| Automation | Desktop and local, no pipeline path described | Batch rendering, the Player for embedding, Lambda or Cloud Run | Coa CLI in scripts, cron jobs, and CI |
| Cost model | No MCP price announced, and the apps must be installed and licensed | Free up to 3 people, then from $25 per seat per month or $0.01 per render | Free forever, or $299 once for the lifetime license. No credit system, no account |
| Maturity | Established suite, brand-new MCP preview | Established, with extensive docs and a public showcase | Young product |
The table shows the features. One question outweighs any single row.
How Adobe MCP, Remotion, and CoAnimator Work
Ask yourself what your video's source of truth should be. Revision cost and who can edit follow from the answer.
- Adobe MCP: A native file in a pro app, edited by an agent and by you in the same project.
- Remotion: Code, where the video is a program and the agent is the programmer.
- CoAnimator: A studio project, where the agent builds scenes and you direct them on a live stage.
None of the three answers is wrong. If you already cut in Premiere, Adobe's route will fit when it opens. If you want video inside your product or rendered from data, Remotion is built for that. If you want demos, launches, and tutorials built from a description and kept as editable files, CoAnimator is built for that. I would match the answer to the team that will maintain the videos. One brief makes the differences concrete.
One Brief, Three Workflows
Take a brief like this one. It is a 60-second video announcing a new billing page, with a revision due a week later. Two habits help in any of the workflows. Give the agent the audience, tone, length, and key moments, and ask for a plan before the build.
Adobe route (preview, not yet runnable)
Adobe's own example is a brief pulled from Google Docs, work done in its apps, and the result shared in Slack. A prompt in that spirit would look like this once the beta opens.
Using the brief in Google Docs, build a 60-second announcement for our
new billing page in my Premiere project, then share the cut in Slack.
The agent edits the native sequence while you watch, and you can step in at any moment.
Remotion route
/remotion:remotion-create Make a 60-second announcement video for our
new billing page. Show the real UI as animated cards, add captions, and
end on the pricing screen.
The agent writes a React composition, you preview it in Remotion Studio, and then you render an MP4.
CoAnimator route
Research how product announcements like this perform on YouTube,
using keyword and trend data. Then research our billing page, draft
a 60-second script, and recommend a length and hook for the platform.
Show me a plan first. After I approve it, build the announcement
scene by scene with the real UI, animated captions, and the pricing
screen at the end.
Here the agent does more than build. Through MCP tools, it can look up what people search for, check how a topic is trending, and recommend a hook and length for the platform before it writes a line of script. Then it builds the scenes while the stage updates. Treat any volume or trend number as research input, and check it against the platform's own analytics before building around it.
How Each Tool Handles Video Revisions
Now legal asks you to remove the word "unlimited" from the pricing line.
- Adobe MCP: Ask the agent to change the text layer in the native file, or change it yourself by hand.
- Remotion: Change the text in the code, or ask the agent to, then re-render.
- CoAnimator: The agent edits the affected scene in the project, and you re-render.
The revision shows where you edit, whether that is a native Adobe file, a line of code, or a scene in a project.
Automating Video Rendering and Releases
Now suppose you want the same announcement to rebuild on every release.
- Adobe MCP: Adobe describes a local desktop workflow and mentions no pipeline path.
- Remotion: Render from data at volume with its CLI and your own infrastructure.
- CoAnimator: The Coa CLI renders without opening the app and exits 0 or 1, so a failed render fails the build.
coa render my-demo --preset 1080p60 --runners 8
coa render launch-video.coa --out launch.mp4
Revisions and rebuilds are where the three workflows diverge, and where cost starts to matter.
Adobe MCP vs Remotion vs CoAnimator Pricing
The three tools charge in very different ways, which changes how revisions feel.
Adobe MCP
Adobe has announced no price for the local MCP integrations. The apps must be installed and licensed on your machine, so your Creative Cloud plan is part of the cost, and your own agent subscription is separate. Adobe's own AI assistants already draw on generative credits, and Adobe says it is exploring upfront cost estimates.
Remotion
- Free License: $0 for individuals and companies of up to 3 people, with commercial use included.
- Creators: $25 a month per seat for companies of 4 or more, covering manual video creation and local motion design systems, including with AI agents and Remotion Studio.
- Automators: $0.01 per render with a $100 monthly minimum, for batch rendering and automated video products.
- Enterprise: From $500 a month.
Remotion's license covers the software only. Your coding agent subscription and any rendering infrastructure you run are separate costs.
CoAnimator
- Free: Free forever, with no expiry and no card.
- Lifetime license: $299 once, tax included, which the pricing page lists as a limited-time price. It includes one year of updates, then updates are an optional $79 a year, and the version you own keeps working.
- Other costs: There is no credit system and no markup. You bring your own coding agent and API keys for plugin services, and those providers bill you directly.
With the cost models clear, the remaining question is what each tool cannot do.
Limitations and Risks
These are the limits I would check before you commit any of them to a team.
Adobe MCP
- The local MCP is not available yet. The private beta starts in early 2027.
- Adobe's apps must be installed and licensed on your machine, because the agent works locally in your project files.
- Accountability is an open question. Fast Company asks who is responsible if an outside agent corrupts a sequence or burns through credits, and Adobe pointed back to the preview status.
- Adobe describes a local workflow and mentions no CI path.
Remotion
- You maintain a React and TypeScript project, even when the agent writes it.
- Teams of 4 or more need a Company License.
- Automated rendering adds a per-render license fee and your own infrastructure costs.
- Good results can take many prompts, and one developer reported multiple days for a launch video.
- Your coding agent subscription is a separate cost.
CoAnimator
- Best results depend on the connected agent, so quality varies with the agent, the prompt, and how much direction is given.
- CoAnimator has no built-in analytics, so tracking who watched a demo must be built separately.
- Export is MP4 only today, and there is no hosted render service, so rendering happens on hardware you control.
- The packaged Coa CLI installer is macOS-first, while Windows renders locally inside the app.
- Premium plugins and research mode, which runs on MCP tools, are part of the lifetime license, and standard plugins on the free plan come with some limits. Connected services use your own API keys.
- The app is still pre-1.0, so expect change between versions.
The limit that matters most for each is availability for Adobe, upkeep for Remotion, and agent dependence for CoAnimator.
Which AI Video Tool Should You Choose?
If you edit in Premiere every day, Adobe's beta is the one to watch. If video belongs inside your product, Remotion is the fit. If you need demos, launches, and tutorials built from a description today, CoAnimator is the fit. I would judge each one on how it handles the second revision, not the first prompt. These five questions turn that into a decision about your own videos.
- Do you already cut in Premiere or After Effects every day? If yes, watch Adobe's beta in early 2027.
- Does your video need to live inside your product or render from data at volume? If yes, choose Remotion.
- Will your video change with every release? If yes, favor a workflow where a change is one edit, such as a scene in CoAnimator or a layer in an Adobe file, over changing code. If the structure repeats and only data changes, Remotion fits well.
- Who will make your videos? With a mixed team of developers, marketers, and founders, CoAnimator lets everyone describe the video while developers still get files they can commit. Adobe asks for Adobe app knowledge, and Remotion asks the team to be comfortable with React.
- Do you need it working today? If yes, Remotion and CoAnimator are the two you can run now. Adobe's cloud connectors work today, but they do not edit your native project files.
This is a workflow judgment, not a benchmark result. Because no source compares the three directly, run a short test yourself. Build a 15-second release video in each tool you can run, then change one line of the script. Record the setup time, the number of prompts, and how long the change took.
-
Remotion: In a new project, run
npx skills add remotion-dev/skills, or install the Claude Code plugin shown above, and prompt for a 15-second explainer. -
CoAnimator: Install the desktop app, connect the agent you already use, open a template or describe the video, then render from the app or with
coa render. - Adobe: Nothing to test locally until the private beta opens. Adobe's own Creative Agent is in beta in its desktop apps if you already subscribe.
If one tool made the change easy and another made it painful, you have your answer.
FAQ
Is Adobe's MCP available now?
Partly. Adobe's cloud connectors for Claude, ChatGPT, and Gemini work today. The local MCP integrations that edit native project files are a preview, with a private beta planned for early 2027.
Does Remotion work with Claude Code?
Yes. Remotion publishes official Agent Skills and a Claude Code plugin, so Claude Code can write the React project and render the MP4.
Do I need to know how to code to use CoAnimator?
No. You describe the video, and the agent writes the project files. Developers also get git-friendly files and a CLI.
Can CoAnimator research topics, keywords, and trends?
Yes. Research mode runs on MCP tools and covers keywords, trends, and SEO checks. It is part of the lifetime license, and you should confirm any figures in the platform's own analytics.
Which one should I pick?
Adobe if you edit in Premiere and can wait, Remotion for video inside a product or at volume, and CoAnimator for demos, launches, and tutorials you want today.
The Verdict, and What to Build First
The result is a split by job. Adobe is the one to watch if you edit in Premiere, Remotion is the fit for video inside a product, and CoAnimator is the fit for demos, launches, and tutorials built from a description. Adobe's move also means agents will soon edit inside pro apps, which makes where your video's source of truth lives the question to settle first.
So pick one recurring video, such as a release recap, and build it in the tools you can run today. I would time the setup and one revision, check Adobe's beta notes when the private beta opens in early 2027, and keep the workflow that makes the second revision easy.

Top comments (1)
Insightful 💡