Describe your idea in a single sentence. Get a full production on an infinite canvas: images, video, voice, music, and every scene wired up for you.
Every creator knows the feeling. You have an idea for a short film, a campaign, a storybook, or just something that lives vividly in your head, and then you sit down to make it and you're suddenly juggling five browser tabs, three apps, and a pile of folders named "final_v7_REALLY_final." The tool gets in the way of the thing.
We built Factory to solve that. It's an infinite creative canvas for AI media, where text, images, video, voice, music, and entire scenes live as nodes you can wire up however you like. At its heart is the feature we're most excited about, Agent Director, which turns one sentence into a fully structured, fully editable production.
You can try Factory today and see it build your first project in minutes.
One sentence in. A complete production out.
Agent Director is the fastest way to go from idea to a finished canvas, and it does considerably more than turn a prompt into a video. It runs an end-to-end agentic pipeline that plans your project, picks the right models for every shot, generates every asset, and wires them all together, in a transparent three-step flow you stay in control of throughout.
In practice, it looks like this. Start with your prompt:
"A 30-second energy drink ad set in a multiverse where Outer Space meets Formula 1. A driver suits up, the can hits the track, and the race tears across a futuristic circuit anchored to a primary vehicle. Cinematic, high-energy, layered audio."
Step 1: Describe your project
Open Factory, click New Project, and choose the Agent Director tab. Paste a brief, whether a sentence, a paragraph, or a rough pitch. You can attach up to eight reference assets and tap Enhance to let Factory tighten the prose before generation. No shot list, no storyboard, no settings page.
Step 2: Review the plan
Before generating anything expensive, Factory writes you a plan. It analyzes your brief, estimates the project scale, picks an approximate node count, and lays out concrete deliverables, shot by shot and asset by asset, for your review.
For the Space F1 brief, the planner proposed a text node with the master campaign script detailing the multiverse transitions and F1 racing beats, two image nodes for the futuristic vehicle concept and a high-end hero shot of the can, a voice node for a commanding announcer track, a cinematic high-energy score, and a sequence of Seedance 2.0 video clips covering the suit-up, the launch, and the race itself. Concatenation and cut nodes stitch everything into a single Final Output.
This is where Agent Director pulls away from the usual "magic button" approach, because you see and edit the plan before the first image renders, swapping models, adding scenes, cutting shots, or tuning the arc. If something isn't quite right, hit Enhance Plan and iterate, and when it is, click Next: Layout.
Step 3: Your project builds itself
Factory lays out every node on the canvas and runs the full workflow. Text nodes write the prompts, image nodes generate references, and video nodes render clips wired to their references and prompts. A voice node performs the narration while a music node scores the bed, and concatenation and cut nodes stitch it all into a Final Output you can play, edit, or export.
Crucially, nothing is locked. Every prompt, every model choice, every reference image, and every clip is a regular node you can tweak, regenerate, or replace, so if Scene 3 isn't landing you can regenerate just Scene 3, swap the music node for a different direction, or branch the tail of the graph into an alternate ending. The agent gives you the first pass; everything after that is yours to shape.
Supervisor and Automation modes
Two modes let you decide how hands-on to be. Supervisor pauses at each step so you can approve prompts, swap models, or rewrite scenes, which matters most when voice and style are doing the heavy lifting. Automation runs end-to-end and brings you a finished draft to react to, which is usually the right choice for a fast first cut.
You can switch mid-project, and a lot of our beta users do exactly that, starting in Automation and flipping to Supervisor for the scenes that carry the emotional weight.
One canvas, every medium
Agent Director works because of what sits behind it: a canvas that treats every medium equally. Right-click anywhere and the node picker appears, with Text, Image, Video, Voice, Music, Storybook, and Infographic, and because each node knows what it makes and what it accepts, connections just work. Image outputs feed video nodes, text prompts feed images, and voiceover syncs to clips without any glue code in between.
Because the canvas is infinite, you can keep every iteration beside your main branch instead of throwing alternates away. Line up four versions of the opening shot and compare them in place, and the iteration stops feeling like a commitment, which is how you start playing with the work again instead of guarding it.
Any-to-Any Context
Every node and every group on the canvas is instantly referenceable, acting as context to inform the generation of new nodes. An image can inspire a song, a voice clip can inform a video, and a whole scene can guide a storybook page, which means your creative assets are never dead ends; they're always inputs for whatever you make next.
Best-in-class models, all in one place
No single model wins every job. A photo-real product shot wants something different than a painterly storybook page, and a cinematic hero cut wants something different than a fast social loop, so we put the best frontier models on the canvas and let Agent Director route to them for you, or let you pick by hand.
- Images: NanoBanana Pro, Seedream 4.5, GPT Image 1.5, FLUX 2 Max, Qwen Image 2 Pro, and Grok Imagine Image.
- Video: Seedance 2.0, Kling O3 Pro, and Veo 3.1.
- Voice & Music: studio-grade voice generation for narration and dialogue, plus original music and score generation.
Swapping models is a dropdown, with no new tabs, no new billing, and no new API keys to wrangle, so you can run the same shot through three models and keep whichever one wins.
Editing that keeps you in flow
Generation is only half the job; the other half is making what you've generated match what's in your head. The image editor handles that without forcing you to start over, so you can swap a background, re-age a character, remove a prop, or restyle a scene for a different hour of day using targeted prompts that preserve character and style consistency across the rest of the frame.
The video timeline is a proper non-linear editor inside the canvas, with tracks, clips, an inspector, cut and split, and precise scrubbing, and it accepts generated clips, uploaded footage, or both.
Video magic tools
On top of the timeline, Factory adds tools built specifically for AI-native video. Multishot generates matching coverage of the same scene (wide, medium, close-up, reverse) from a single description, motion control lets you steer camera moves and subject motion through plain language, and consistency controls keep your lead in the same outfit and your setting on the same street from scene one to scene ten.
When you're ready, hit Production to render a final film, with cuts, audio, and color, directly from the canvas.
Stories, spelled out page by page
Some ideas want to be films and others want to be picture books or manga chapters. The Storybook node is a dedicated workflow for multi-page illustrated stories, where you feed it an outline and a few reference images and it generates every page with consistent characters, art style, and tone. It's a good fit for bedtime stories, illustrated pitch decks, brand mascots, or an afternoon of character experimentation.
Made to be shared
Projects in Factory aren't files you email around. Every project has real-time collaboration, comment threads, and owner/editor/viewer permissions, so you can invite your team and work on the same canvas at the same time, or hand off a draft to a producer and let them leave notes directly on the nodes.
You can also make a project public. Public projects appear to the community, who can view the canvas, see exactly how it was built (every prompt and model choice visible), and clone it into their own workspace to riff off, which is closer to sharing a recipe than sharing a meal.
What you can actually build with it
Short films, animated shorts, and music videos. Product shoots, fashion look books, and campaign visuals. Explainer infographics, pitch decks with original narration. Picture books and manga chapters. Album art, loop-ready motion graphics, launch banners. Agent Director turns "a week and four freelancers" into "a first draft before lunch," and leaves room to refine it into something personal.
Start building
Factory is live at agent.ii.inc/factory
Bring an idea, paste a sentence, or just try Agent Director and watch a full production assemble itself. The canvas is ready when you are, and we'd love to see your project when you ship it.
Keep up to date with us on X for new features, beta invites, and builds from the community.