FLUX-mimic
#21 in Robotik & Embodied AIMimic Robotics · 2× · tolest 25. Juli 2026
FLUX-mimic is a video action model for robotics that mimic robotics developed together with Black Forest Labs based on the FLUX-3 video model. Rather than building on classical Vision-Language-Action architectures, it leverages a pretrained generative video model that has learned world dynamics from large-scale video material, and trains a compact Action-decoder on top that derives robot actions from predicted visual futures. According to the provider, this enables new manipulation tasks to be learned with significantly less robot data (in some cases from 30 minutes instead of previously 30+ hours). The model has been tested and deployed at Audi since July 2026 in production and logistics environments, including for complex soft-body manipulation.
Features
| Deployment Model | Gated early access for selected partners (no public availability); shipped as part of FLUX 3 Action, per secondary source deployable on a single on-prem GPU |
| Use Case Scope | Industrial robot manipulation: kitting parts, inserting electronic control units, assembly, handling soft/flexible materials (e.g. seals, cables) in production and logistics |
| Integrations | First industrial deployment partnership with Audi (testing and deployment on the factory floor); developed in collaboration with Black Forest Labs (FLUX 3) |
| Platform | Video-Action Model built on Black Forest Labs' FLUX 3 video backbone, with an additional action decoder for robot control |
| Release Date | July 23, 2026 (announced/preview alongside FLUX 3) |