A creative production studio, built on a node canvas, where spatial ideation meets real execution.
Drag nodes onto an infinite canvas, wire them together, and hit run to generate images, video, voice, music, 3D, and talking-head performances without leaving the workflow. Then version it, review it with your team, and cut it in a built-in editor. An orchestration layer for generative work, built for a solo creator or a whole studio.
Or just talk to the built-in AI agent, and watch it build and run the entire workflow in front of you.
Where it fits
And Action! doesn't change how your artists work. It coordinates the generative layer around them: the single place where generative workflows are planned, run, versioned, and reviewed.
The AI agent · your collaborator
Describe what you want in plain language, and the agent constructs the node graph in front of you: wiring nodes, choosing engines, running the graph, and suggesting the next move as it goes. It isn't a chatbot bolted onto the side. It operates the real application, through the same 55 tools and the same canvas you use.
Illustrative: a request on the left becomes a live, running workflow graph on the right, built by the agent and refined with you.
New to the canvas? Ask “how do I…?” and it answers from the product itself: it points you to the right node or panel, or just does it for you. The fastest way to learn the app is to watch it work.
It does anything you can: builds multi-step pipelines, runs and verifies jobs, cuts on the timeline, files review notes, forks and tracks assets, searches across every project. A power user that never tires of the boilerplate.
New users learn the craft by watching the agent work. And the agent learns you, remembering per-project preferences over time, so its choices and suggestions sound more like your studio the longer you work together.
Because it drives the real app through an open tool surface (MCP), the same agent is reachable from external clients too; it's the product's operator, not a black box bolted on.
The core loop
The canvas is home base: a spatial workspace built for rapid iteration without sacrificing control, provenance, or collaboration.
Arrange nodes, wire them together, hit run. Results stream back live onto the canvas.
Instant previews for iteration. When you're happy, save a Take: a versioned, permanent record of the exact workflow and inputs that made it.
Pick the best model per node. Mix providers in one graph; the app auto-detects what's available and free.
The canvas
Assemble complex creative pipelines across nine palette groups (generation, mood board, inputs, filmmaker, analysis & simulation, workflow utilities, and publishing) without writing code.
Automatic dependency ordering (topological sort), live per-node progress streaming, cascade re-runs when you change an upstream node, and full run history with timing and output links.
Independent branches of a graph execute concurrently on "Run to," so a wide board finishes in the time of its slowest chain, not the sum of all of them.
Pen, shapes, text, stickies, and frames alongside your nodes. Free rotation, frame aspect presets + numbering + ordering, comment pins, storyboard templates, and camera-move glyphs.
Presentation mode plays through ordered frames; the animatic assembler stitches them into a cut; contact-sheet export renders the board to a single sheet.
Generate anything
Each generation node offers per-node engine selection; runtime detection auto-picks the best available. Mix and chain freely: generate → edit → upscale → interpolate → grade.
| Engine | Best for |
|---|---|
| Gemini free tier | Fast iteration and quick experiments |
| Seedream 5 (Pro / Lite) | High-detail text-to-image and prompt-guided edits |
| FLUX Pro (Fal) | Professional-grade quality |
| GPT Image 2 (OpenAI) | Detailed generation and mask-optional edits |
| Luma Photon | Photorealistic imagery |
| Pollinations free | No-key generation to start instantly |
Plus image editing (i2i modify), expand (outpaint beyond the frame), upscale (free locally, or Topaz), local color grade, and transform (crop / resize / rotate / flip). Grade, transform, and the default upscale run free on your machine via FFmpeg and Pillow.
| Engine | Capabilities |
|---|---|
| Sora (OpenAI) | Generate, extend, in-paint, character consistency |
| Luma Ray | Generate, modify (9 modes), extend, 4K upscale, reframe, interpolate |
| Runway Gen-4 | Text- and image-to-video |
| Kling · Wan · Hailuo · Seedance (Fal) | A range of generation styles and speeds |
| Netflix VOID | Object removal & video inpainting (heavy model, dispatched to cloud GPU) |
LongCat-Video-Avatar drives a photo from an audio track into a lip-synced performance, single or multi-character.
TRELLIS.2 turns a still into a 3D asset for downstream use.
SAM2 predicts masks on images and video: the backbone for clean object removal and rotoscoping.
| Feature | Options |
|---|---|
| Voiceover | ElevenLabs, Fish Audio (80+ languages, emotion control), Edge TTS free, local F5-TTS local |
| Music | MusicGen and hosted generation models |
| Sound design | FFmpeg local · free |
| Captions | Whisper transcription local + burned subtitles |
Use language models as pipeline components: Claude, GPT, Gemini, and Grok for reasoning, ideation, scripting, and visual analysis; Ollama local · free for always-available offline processing.
Cinematic control
Filmmaker-grade controls to compose shots for previs or final delivery.
Built for teams
Collaboration is a first-class feature, not an afterthought, designed for remote teams across timezones, and to fit a studio's existing pipeline rather than become another silo.
Server-authoritative shared editing with presence, remote cursors, per-user undo, and reconnect resync. Two people can shape the same board at once.
Stream and scrub footage frame-by-frame, draw annotations in real time, attach structured notes to specific frames, and build playlists across takes, all without an external viewer.
Pin comments to the canvas and thread feedback where the work lives, so context never gets lost in a side channel.
Every saved take flows into a versioned catalog searchable across projects, with full lineage back to the workflow that made it.
Self-hostable, with role-based access and admin controls. Studios administer the instance; artists get full functionality, including the agent.
API-first and database-canonical, designed to interoperate with your existing DAM and review systems over its API: orchestration that plugs in, not a wall you route around.
Production pipeline
Previews are transient; Takes are permanent, versioned assets. Each one snapshots the exact graph and resolved parameters that produced it, so any take can reopen the workflow that made it back onto the canvas months later, deprecated nodes and all. Promote the best version into the Studio Library, a disk-canonical snapshot pin that survives project archive, offline, and hard-delete, so a shipped asset can never quietly disappear. And a workflow is itself a first-class versioned asset: publish it, version it, and reuse it like any other output.
A full non-linear editor lives inside the app, with 84 AI tools reachable by the agent across timeline, clips, effects, masks, keyframes, and transitions.
Edit directly from takes, and chain editor output back onto the canvas for further generation.
Under the hood
Heavy models that can't run on consumer hardware are dispatched to cloud GPUs (Modal, Runpod) through a provider-agnostic layer, with a budget gate that stops billing before it starts.
Per-job spend is tracked to a persistent ledger with a monthly cap and live spend pill, plus per-engine run/success/cost telemetry, all surviving restarts.
Projects, takes, notes, published assets, and canvases are database-canonical with disk projections: reliable state, not a pile of JSON files.
Ollama, FFmpeg, Whisper, Pillow, and stock media run locally at no cost. API keys are optional and only needed for premium engines.
Status
And Action! is in active private beta. New engines, nodes, and features land regularly.