FLUX 3 Action Base
FLUX 3 Action Base is a 7B world-action adaptation component. It jointly denoises future visual frames and action sequences from camera observations, robot state and a text instruction.
Model details
The package contains base action weights and shared frozen encoders, including Qwen3-VL-4B-Instruct and a video VAE. It has video and text streams, a shared action trunk and existing action heads. A new robot embodiment requires its own action head.FLUX 3 Action Base official documentation
Practical use
Practical guidance. Use the base when adapting the model to a new robot, simulator or game. Define the observation and action units before training. A plausible predicted video is not proof that the motor commands match the hardware. Validate the complete control contract in simulation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | black-forest-labs/flux-3-action-base |
| Weight license | FLUX Kommunity License 1.0; text encoder separately under Apache 2.0.FLUX 3 Action Base official documentation |
| Role | Adaptation component; not a complete ready-to-run robot policy.FLUX 3 Action Base official documentation |
Related models
- FLUX 3 Action DROID
Black Forest Labs · FLUX 3
Seven-billion-parameter robot policy that generates joint commands and future frames from DROID-style observations.
Launch date not confirmedDownloadable Seven-billion-parameter SO-101 robot policy that predicts actions and future frames from cameras, robot state and task instructions.
Launch date not confirmedDownloadable