← Back to blog
AiAbout 6 min read

YouArt Puts an Agent Inside After Effects, and Hands You the Model Bill

Published Nov 14, 2026
YouArt Puts an Agent Inside After Effects, and Hands You the Model Bill

The demo that sells YouArt Desktop never shows a generated clip. It shows a client note. Someone writes in Slack that the product shot at 00:07 should be tighter, and the app pulls that message into a project canvas as a comment pinned to that shot. The agent then goes into the edit, makes the change, and leaves the revision for a human to accept or throw out.

YouArt, a Y Combinator-backed company, launched the Mac and PC app on November 13. Its pitch is narrower than "AI makes your video" and more useful: the agent works inside the software you already cut in. With permission, it operates Adobe After Effects, Premiere Pro, DaVinci Resolve, CapCut and Blender by navigating menus, clicking, and typing commands. Your project files, templates and plugins stay where they are. Nothing gets imported into a walled garden.

What it actually touches

The canvas is the part worth understanding. It holds footage, scripts, brand assets and generated clips, plus comments and version history, so the agent and the editor look at the same objects. A note from Gmail or Slack becomes an annotation tied to a shot, a frame or a single element, and the agent reads that annotation as the instruction.

From there it can do production work: turn product assets and brand material into a storyboard, build visuals and motion graphics, animate them inside the app, then export to After Effects for the finishing pass. YouArt also ships reusable skills for narrower jobs like story development, shot planning, commercial video production and motion design, which is the same pattern every agent product landed on this year: a general worker plus a shelf of recipes.

There is a Chat to Edit mode for the small stuff. Select a title, say "use our brand blue" or "hold this for one more second", and the agent applies the change while leaving the rest of the timeline alone. Manual controls stay available the whole time, which matters more than it sounds. An agent that can only act through conversation is useless the moment you need to nudge a keyframe by four frames.

The app also borrows an idea from browsing agents: built-in browser tabs for research and references, with a visible indicator when the agent is driving a page and a way to take the wheel back.

The pricing decision is the interesting one

Every AI video platform of the last two years bundles model access into a subscription. You buy credits, you spend them, and you never quite know what a generation cost. YouArt goes the other way. You bring your own API keys and connect third-party providers, including Higgsfield, fal and BytePlus. A routing layer compares prices across providers and picks the cheapest option for whichever model you selected.

That is a small feature with a large consequence. The moment a creator can see that the same model costs different amounts at two providers, the platform stops being able to hide a margin inside credits. It also changes how people work: a draft shot gets routed to the cheap endpoint, the final frame gets the expensive one, and the decision is made per generation instead of per month.

The rest of the stack follows the same logic. Voiceover runs through ElevenLabs v3, music through Suno, and lip sync and avatars are in the same workflow rather than in three other tabs. AutoClip takes long footage and cuts it into short clips with captions and per-platform formatting, which is the task most marketing teams actually need and least enjoy.

Blender support is the detail that reframes the product. Editing suites are one category; a 3D application is another, with a scene graph, a node-based material system and a render pipeline that punishes wrong clicks. Letting an agent drive Blender means it can place a product model into a shot, adjust a camera path or render a turnaround without a 3D artist being booked for the afternoon. Whether it does that reliably is a different question from whether it can, and the launch material does not settle it.

Three products, one week, same argument

YouArt shipped two days after Kino AI opened its MCP-based editor, and one day before DomoAI released InkoAI Agent. Three companies, three architectures, one conclusion: the agent does the steps and a person approves the output.

Kino exposes the media library and timeline as an MCP server so Claude Code, Codex or Cursor can edit footage from their own interfaces. DomoAI runs the whole pipeline in one chat across 15 image and video models, with approval gates at script, character and shot. YouArt sits closest to existing professional work: it drives the NLE itself rather than replacing it or wrapping it in a protocol.

That difference decides who each product is for. A team with ten years of After Effects templates will not move to a chat window. It might let an agent open the project and do the first assembly pass. A solo creator with no templates is better served by DomoAI's one-sentence-to-clip route, because there is nothing to preserve.

Where this breaks

Driving a GUI is the least reliable way to control software. Menu positions shift between versions, plugins throw dialogs, and an agent that clicks by coordinates will eventually click the wrong thing. YouArt's answer is permissions plus review: the agent acts, the creator checks, nothing ships unread. That is honest, and also a confession that the automation is not trusted to finish.

Cost is the second risk. Bringing your own keys means bringing your own bill, and an agent that iterates without asking can run through a lot of generations. Smart routing softens it, not eliminates it. The teams that benefit most will be the ones that set a budget per shot before the agent starts, which no vendor can do for them.

The third is responsibility. If a frame is wrong, the question goes to whoever pressed approve. That has not changed since the first AI video tool shipped, and no amount of menu automation moves it.

Who should care

Marketing teams with a product launch, a brand system and no appetite for a new editor. Post houses that want a first assembly before a human editor starts. Filmmakers doing storyboards and pre-vis on their own machines. All three have something in common: existing assets, existing templates, and a fixed idea of what the output should look like.

For everyone else, the interesting part of YouArt's launch sits in the price list rather than in the agent. Once a platform lets you see what each generation costs at each provider, the conversation about AI video pricing changes from "how many credits" to "which model, at which price, for which shot". That is a better conversation, and it took a desktop app to force it.

Related articles

Google replaced the model behind Search, and the number that moved was a shell benchmark

Gemini 4 Hydrogen became the default on 12 November, taking over from Gemini 3.8 Flash across Search, the Gemini app and the developer API. It runs about 2.5 times faster, scores 57% on Terminal-Bench 4.1 where its predecessor scored 18.2%, and costs $1.50 in and $7.50 out per million tokens.

Three ways a coding agent wastes money, measured on 30 repositories

TokenEconomizer watches what an agent did in a repository you work in often, then stops it from rediscovering the same build command, loading tools it never uses and looping on one error. Across five agent and model combinations it cut mean cost by 18.4% to 30.9%, cheaper in all 30 repository cells, with Pass@1 up in four of five.

earshot gives coding agents 922 sounds, and every one is rendered from code

An agent can write a game, an interface or a film and never hear any of it. earshot answers requests like "three variations of a footstep on gravel" from a catalogue of 922 designs built from public-domain recordings, with no audio files in the repository at all. It runs on CPU, as a CLI, a Python package, an MCP server and a Claude Code plugin.

The agents stealing API keys were not sophisticated. The gateways were open.

Amazon's decoy servers spent October posing as inference APIs and logged agents built on ARTEX pulling provider keys out of LiteLLM, Dify, Langflow and Open WebUI. The way in was a default master key and a half-finished install. ARTEX itself ships with budgets, scope gates and evidence for every finding.