> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orchestrallabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# FLUX.3 Action

> Inspect and deploy FLUX.3 Action, with its camera and state inputs and action output.

## Checkpoint

```bash title="Your computer" sample=verified theme={null}
orchestral models inspect flux3_action hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4
```

```text title="Sample output" theme={null}
Inputs:
  images.wrist: RGB uint8 (360, 640, 3)
  images.left: RGB uint8 (360, 640, 3)
  images.right: RGB uint8 (360, 640, 3)
  state: float32 (8,)

Outputs:
  actions: float32 (32,8)
```

## Inputs and output

Use the camera keys listed below. Missing or extra keys raise an error.

| Direction | Name | Shape | Type |
| - | - | - | - |
| Input | `images.wrist` | `(360, 640, 3)` | RGB `uint8` |
| Input | `images.left` | `(360, 640, 3)` | RGB `uint8` |
| Input | `images.right` | `(360, 640, 3)` | RGB `uint8` |
| Input | `state` | `(8,)` | `float32` |
| Output | `actions` | `(32, 8)` | `float32` |

## Deploy and predict

Deploy the checkpoint URI directly. Orchestral resolves, imports and prepares it before bringing up the serving deployment.

```bash your computer theme={null}
orchestral deployments create hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4 --robot robot-1 --wait
```

```text title="Sample output" theme={null}
jolly-tangerine-snail-rdik  ACTIVE · ready
FLUX.3 Action · robot-1
Timeline: requested 12s ago → starting worker 8s ago → ready now
```

```python robot computer theme={null}
import numpy as np
import orchestral as ol

client = ol.Orchestral()
deployment_id = input("Deployment ID from the CLI output: ").strip()
deployment = client.deployments.get(deployment_id)

inputs = {
    "images.wrist": np.zeros((360, 640, 3), dtype=np.uint8),
    "images.left": np.zeros((360, 640, 3), dtype=np.uint8),
    "images.right": np.zeros((360, 640, 3), dtype=np.uint8),
    "state": np.zeros((8,), dtype=np.float32),
}

with deployment.session() as session:
    result = session.predict(inputs=inputs, instruction="pick up the object")

print(result.actions.shape)  # (32, 8)
```

## Imported policies

Inspection reads the checkpoint’s upstream `PolicyConfig`, including camera keys, layout,
action width, and chunk size. Orchestral calls the upstream policy’s preprocessing and output
transforms, including configured normalization and action conversions. The default checkpoint
uses the three inputs shown above. Temporal camera-history configurations raise an error.

## Variants

The checkpoint provides the `sd-fp8r` and `base` variants. Choose one explicitly when you need
to pin a particular variant:

To pin FP8r explicitly during inspection and deploy:

```bash your computer theme={null}
orchestral models inspect flux3_action hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4 --options variant=sd-fp8r
orchestral deployments create hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4 --options variant=sd-fp8r --robot robot-1 --wait
```

To select the base variant explicitly:

```bash your computer theme={null}
orchestral models inspect flux3_action hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4 --options variant=base
orchestral deployments create hf://black-forest-labs/flux-3-action-droid@e3b6c915cdcdfb1d5db08e133b8ca3d29be2d7d4 --options variant=base --robot robot-1 --wait
```


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.