# AnimGraph

An animation graph turns gameplay values into a character's pose. Your game says "I'm moving at this speed" or "start waving"; the graph decides which clips to play and how to blend them.

**AnimGraph (AG2)** uses `.ag` files. This guide covers the new system, not the legacy `.vanmgrph` editor.

## How a graph works

The simplest graph plays one animation:

```text
Animation Clip (idle) ---> Graph Output
```

An **Animation Clip** produces a pose: the positions and rotations of the character's bones at that moment. **Graph Output** is the pose you want the character to use.

Other nodes can blend poses, choose between them, or adjust individual bones before the result reaches the output. **Parameters** carry values such as speed, grounded state, or an aiming target into these decisions.

There are two kinds of connections to recognize:

| Connection | What it carries | Example |
|---|---|---|
| Pose | An animation result | A walk clip feeding a blend |
| Value | A number, bool, ID, vector, target, or mask | A speed parameter controlling that blend |

Unlike ActionGraph, pose connections are not execution signals. You are describing how to produce a pose, not a sequence of commands. A clip in an inactive state does not keep playing in the background independently.

## Which node should I use?

| I want to... | Start with |
|---|---|
| Play an animation | **Animation Clip** |
| Blend idle, walk, and run from speed | **Blend 1D** |
| Blend movement from two values, such as forward and sideways speed | **Blend 2D** |
| Change behavior while playing, such as idle to wave | **State Machine** |
| Pick a variation once when entering a branch | **Selector** |
| Add a pose over another pose | **Local Layer** and **Layer Blend** |
| Play an action layer with its own states | **State Machine Layer** and **Layer Blend** |
| Restrict a layer to part of the body | **Bone Mask** |
| Share one graph structure across character-specific animation setups | **Variations** |
| Reuse an authored graph driven by parent parameters | **Internal Graph** |
| Let items supply custom holds or actions at runtime | **External Graph** |

**A Selector chooses on entry, not continuously.** If changing a parameter should switch the animation immediately, use a state machine or a blend instead.

For the full list, see [Node Reference](/dev/doc/animation/animgraph/node-reference): editor node names, what each one does, and the important restrictions.

## Learn by building

Start with [your first graph](/dev/doc/animation/animgraph/getting-started), then extend it:

1. [Blend movement](/dev/doc/animation/animgraph/blending-movement) from one speed value.
2. [Add states and transitions](/dev/doc/animation/animgraph/states-and-transitions) to play a wave and return.
3. [Use layers and masks](/dev/doc/animation/animgraph/layers-and-masks) to wave while the legs keep moving.

[Skeletons and clips](/dev/doc/animation/animgraph/skeletons-and-clips) explains asset preparation. [Parameters and code](/dev/doc/animation/animgraph/parameters-and-code) connects the graph to your game.

[Animation Events](/dev/doc/animation/animgraph/events) explains typed callbacks, ID markers, foot phases, native events, and graph events, with clip-authoring steps and a C# receiver.

[Variations](/dev/doc/animation/animgraph/variations) let one graph serve a family of characters or items, with inherited animation data and targeted overrides.

[Internal Graphs](/dev/doc/animation/animgraph/internal-graphs) explains reusable child assets, parameter passing, and shared locomotion. [External Graphs](/dev/doc/animation/animgraph/external-graphs) shows how to extend a main character graph with custom item hold types, draw-to-hold sequences, and runtime-selected actions, using either the editor or C#.

Want to construct graphs in C#? [Runtime Graph Builder](/dev/doc/animation/animgraph/runtime-graph-builder) covers clips, blends, states, layers, procedural animation, and runtime-selected actions.

Coming from the old editor? [Moving from AG1](/dev/doc/animation/animgraph/moving-from-ag1) maps the old workflow to the new one.