Generative work that stays in-house.
Studio and brand contracts increasingly forbid sending client material to public AI services. nooodles runs open models where you choose — so the work you currently have to decline becomes work you can bid.
One package, and none of it leaves the building.
Everything the work needs, in one graph
Modelling and deformers, a cloner and effector stack in 2D as well as 3D, real physics, a dope sheet and f-curve editor with camera cuts and reusable shots, compositing with real blend modes and levels, typography off your installed fonts, live data, and the page it ships on. Nothing gets exported to the next application, because there isn't one.
One package instead of a shelf of renewals
A loaded motion seat is several subscriptions deep — something for 3D, something for motion and compositing, something for the real-time work, something for the web side, and a plugin bill on top of all of it. This is one seat covering the ones a job actually touches, and it renews once.
Built AI-native, not retrofitted
The engine was designed with generation as a first-class operator rather than having a panel added to it later, and that is the whole difference. Generation is one operator wired into what it feeds — a texture pass on a model, a background behind a plate, a variation set under a director's note. Parameters, seeds and inputs are explicit, so a note like "same, but the light comes from the left" is a change you make rather than a prompt you re-roll and hope.
The graph is documented creative direction
A prompt is a wish. A graph is a set of decisions you made, written down, with the human choices legible — composition, timing, materials, what generation was allowed to touch and what it wasn't. US law protects human-authored expression (D.C. Cir. 2025; cert denied 2026), and an explicit authored graph is the strongest authorship position AI-assisted work can hold. It isn't a guarantee of ownership, and we won't pretend otherwise — take the specifics to your own counsel.
The weights stay put. The work never makes the trip.
It runs where you choose — in the cloud on a key you paste, so nobody needs a GPU to start, and on your own hardware when the work can't leave the building. The model runs next to the material instead of the material travelling to the model. Unreleased product, an unannounced identity, a campaign still under embargo: none of it leaves.
The constraint is the contract, not the craft
Major studio partner programmes and the MPA and TPN content security standards are the reason your clients now ask where the material goes and whether anything trains on it. Those are their standards, not a nooodles certification — we're not claiming one. The point is narrower and useful: a tool that can run inside your own secured environment is answerable to that review, and a cloud-only tool structurally isn't.
The usual alternative is standing up private GPU infrastructure yourself — hardware, racking, someone to run it, and a bet on which models matter in eighteen months. That's a board conversation with a payback period. Deciding per project, on the jobs that actually carry the constraint, is a different kind of decision entirely.
Animate it, then change your mind.
A real dope sheet and curve editor, a diamond on every parameter, and a graph underneath that never bakes — so retiming a move or reworking the look doesn't mean animating it again.
Keyframe anything, not just transforms
A diamond sits on every parameter. A shader value, a light's intensity, a physics setting, a material colour, the text in a headline, the mode in a dropdown — if it's a parameter, it takes keys. The things you usually have to fake because they aren't animatable are animatable.
Retime without re-animating
Alt-drag a selection in the dope sheet to squash or stretch a whole move. Reverse a move and the eases reverse with it. Weighted bezier tangents you drag, break and equalise. The client asking for it 20% snappier is a drag, not an afternoon.
Cut cameras like an editor
A camera-cut track where switching cameras is a clip you drag and trim, not a rewiring job. And a Shot is a reusable sub-timeline — place it with an in-point and a speed, drop the same one into two sequences, double-click to open it and change it in both.
Motion that doesn't need keys at all
Oscillators, noise, envelopes and response curves as signal operators, wired straight into whatever they drive. Cloners with effector stacks and fields you drag in the viewport. Procedural where procedural is better, keyed where a key is better, in one timeline.
Type that moves per character
Your installed fonts, edited directly on the render with a real caret. Shatter a headline into per-character glyphs and stagger them with an effector — a reveal that stays exact when you scrub, in one GPU pass rather than 40 layers.
Nothing is baked, so nothing is a round trip
The geometry, the materials, the compositing and the animation are one graph. Change the product model and the move you animated is still on it. There is no render, re-import, re-comp loop, because there is nothing to re-import.
What your people actually get to make.
A procedural toolset your TDs and animators already have the instincts for, without leaving the browser.
Procedural 3D
Deformers that stack and never bake, bevel and extrude, voronoi fracture with per-shard rigid bodies, surface scatter, and fields you drag in the viewport.
2D MoGraph
A cloner and a five-effector stack on the image chain, not only in 3D — so a grid, a scatter or a field-driven layout is the same gesture in both places, in one GPU pass.
A real sequencer
A dockable dope sheet and f-curve editor with weighted tangents, alt-drag retiming, a camera-cut track and reusable shots. Professional grade, and it keys any parameter.
The frames, and the piece that runs.
Today
The brief has a generative component and an NDA that forbids public AI services. You have two options: turn the work down, or commit to building private infrastructure before you know whether the work is repeatable. Most studios pick the first and don't put it in the deck.
With nooodles
It's a project decision. The same graph runs against a cloud model for the jobs that allow it, and inside your own environment for the ones that don't. You bid the work, and where the models run is a line in the technical response rather than a reason to pass.
It pitches the work, not just the frames.
Frames, boards, title variants and reels are a workflow — material your team cuts and places. But the same graph can also emit a scene: a live piece the client opens in the browser and moves through themselves.
Previs a sequence they can scrub. Send a look-dev comparison they can toggle. Pitch an interactive idea as the thing itself instead of forty slides describing it. Same source, same art direction — one output is files for you, the other is something they use.
The delivery
Frames, boards, title sets and cuts — files your team takes into the edit.
- Mode
- Workflow
- Who opens it
- You and your team
The pitch and previs
A live link the client opens and drives — options, timing, look, in the browser.
- Mode
- Scene
- Who opens it
- Your client, in the room
When something works, save it and run it again.
These ship with the beta, and every one of them runs on your own hardware too.
Private Pipeline
WorkflowRun generation inside the client's own environment. An offering we're defining with early customers — not something you can buy yet.
Sequence Frames
WorkflowHold look and character consistent across a sequence.
Configurable Reel
SceneReorder a showreel per viewer or brief.
Title Set
WorkflowProduce motion titles across durations and ratios.
Frame Extractor
WorkflowPull stills and cutdowns out of a timed piece.
Social Cutdowns
WorkflowOne edit, every duration and ratio the calendar needs.
What your pipeline lead will ask.
What does this actually replace?
The parts of a motion job that are procedural: modelling and deformers, cloners and effector stacks in 2D as well as 3D, physics, a dope sheet and f-curve editor with camera cuts and shots, compositing with real blend modes and levels, typography off your installed fonts, and the page or link the work ships on. One seat, renewing once, instead of a subscription for each of those.
What it does not replace: a finishing suite on a hero film, or an edit bay. If the job ends in a grade and a conform, that pipeline stays where it is. This is for the work that has to exist in twelve versions and stay right in all of them.
Will it fit our pipeline?
What comes in: GLB and glTF as geometry you can still deform, layered PSD and Illustrator files as an editable layer graph rather than a flattened image, SVG as real paths, HDRI for lighting, and the fonts already installed on the machine.
What does not: there is no importer for another tool's project or timeline, and building one honestly is not on the near list. You rebuild the setup once — and because it is a setup rather than a comp, the versions after that come out of it rather than being made again.
What exactly runs locally?
The editor is a browser application and the engine cooks on your machine — geometry, materials, motion, compositing and the document itself are local already. The part that has a choice is generative inference: it runs in the cloud on a key you paste, or against open models on hardware you control, so the weights stay put and the client material never leaves your network. Which specific models, on which hardware, is a setup conversation — that's the piece we're defining with early customers.
Do you train on our work?
Three honest answers, because they're different. Today generation runs on a key you hold, so the material goes to your provider under your agreement with them, not to us. On a local or self-hosted setup the question is settled by construction: nothing phones home, because there's nowhere for it to go.
And if we host generation later, we may use graph structure — how nodes are wired, not your source material — to improve the built-in assistant. Written no-training terms are part of the Enterprise conversation rather than a document you can read today, and we would rather say that than point you at terms that aren't written.
Can we get it through our security review?
We'll do the work with you. The shape that usually passes is the local one: the material stays inside your environment, the models run on your hardware, and the data-flow answer is "it doesn't leave." We can't hand you a certification — we don't hold one, and any vendor who tells you their certification satisfies your client's audit is skipping a step. We can hand you an architecture you can put in front of a reviewer, and answer their questions directly.
Which models can we run?
Open models, and the next one. We deliberately don't build the product around a fixed list — generation is a node with inputs and parameters, so a model is something you point it at rather than something the tool is made of. That matters more than which name is fashionable this quarter, because the list changes every few months and your graph shouldn't have to.
Is there a real curve editor, or just eases?
A real one. A dockable panel with a dope sheet and an f-curve editor, weighted bezier tangents you drag, break and equalise, and normalised or stacked views for comparing channels. Per-key easing, and pre- and post-infinity behaviours including cycle-with-offset — so two keys can give you a move that never stops.
How do I retime something that's already animated?
Marquee the keys and alt-drag an edge to squash or stretch the selection. Time-reverse swaps the eases with it, so a reversed move doesn't come back feeling wrong. Because the keys sit on parameters rather than on baked layers, retiming doesn't cost you the look — nothing has to re-render to be correct.
Bid the work you're turning down.
Free to start, nothing to install. Or tell us what your contracts require and we'll walk through what running it inside your environment would actually look like.
Tell us what your contracts won't allow
Local weights, air-gapped work, NDA constraints — the requirements that rule most tools out are the ones we want written down properly. Describe yours and we'll work through what running it inside your environment takes.
No spam. A real reply from a human.