Flux-mimic
#9 v Robotika a Embodied AIBlack Forest Labs · 5× · naposledy 28. 7. 2026
FLUX-mimic is a video-action model for robotics that Black Forest Labs developed together with Zurich-based robotics startup mimic robotics based on the FLUX-3 backbone. It decodes robot actions directly from the learned world representations of the video model, enabling general-purpose dexterity with significantly less demonstration data than classical Vision-Language-Action models. The system is currently being tested in real factory environments in collaboration with Audi, including applications such as kitting, connector assembly, and handling of flexible materials like cables and seals. FLUX-mimic is in a gated Early-Access phase for selected research and industry partners; public pricing or an API does not yet exist.
Vlastnosti
| Deployment Model | On-premises deployment on a single GPU in factory environments; fine-tuning for new tasks possible with about 30 minutes of robot data |
| Use Case Scope | Industrial manufacturing/logistics: kitting parts into trays, inserting control units into fixtures, assembly, handling flexible materials like seals and cables |
| Integrations | Integrated into mimic robotics' full-stack deployment system (robot hardware, sensors, actuators); tested and deployed at Audi |
| License | No open license; not open-source, closed model with gated partner access (FLUX 3 Dev open-weight release planned for later 2026, primarily for content generation) |
| Platform | Runs on a single on-premises GPU (tested on NVIDIA RTX 5090); backbone processing under 80ms, end-to-end reaction time ~101ms |
| Price | No public pricing; gated access for selected partners only |
| Release Date | July 23, 2026 (announced as 'FLUX 3 x mimic') |