--- title: flux-3-action-so101 canonical_url: "https://www.modelscope.cn/models/black-forest-labs/flux-3-action-so101" md_url: "https://www.modelscope.cn/models/black-forest-labs/flux-3-action-so101.md" repository: black-forest-labs/flux-3-action-so101 last_updated: 2026-10-02 license: other pipeline_tag: robotics tasks: - robotics base_model: - black-forest-labs/flux-3-action-base base_model_relation: finetune parameters: 6.9B tensor_type: - F32 - BF16 library_name: - safetensors - pytorch frameworks: - pytorch downloads: 112 stars: 7 tags: - robotics - lerobot - so101 - flux - world-action-model - world-model - robot-learning - black-forest-labs - flux-3 --- # flux-3-action-so101 > flux-3-action-so101 - black-forest-labs 在 ModelScope 开源的模型。FLUX 3 Action SO-101 black-forest-labs/flux-3-action-so101 是 ModelScope 魔搭社区上的 6.9B 参数robotics模型，采用 other 许可，基于 black-forest-labs/flux-3-action-base 构建。 - **Repository**: black-forest-labs/flux-3-action-so101 - **License**: other - **Tasks**: robotics - **Parameters**: 6.9B - **Base model**: black-forest-labs/flux-3-action-base - **Tags**: robotics, lerobot, so101, flux, world-action-model, world-model, robot-learning, black-forest-labs, flux-3 - **Downloads**: 112 - **Stars**: 7 - **Last updated**: 2026-10-02 Source: https://www.modelscope.cn/models/black-forest-labs/flux-3-action-so101 --- # FLUX 3 Action SO-101 FLUX 3 Action is an open weights 7B world action model. It takes camera frames, the robot's state and a text instruction, and returns the next chunk of actions, denoised together with the next video frames. This repository is part of the [FLUX 3 Action collection](https://huggingface.co/collections/black-forest-labs/flux-3-action-6ab25aef555dd30ab86567f8). For more information, read the [documentation](https://docs.bfl.ai/flux_3/flux3_action_overview). The training and inference code is in [black-forest-labs/flux-action](https://github.com/black-forest-labs/flux-action) on GitHub. SO-101 policy with checkpoint-owned observation history, camera layout and normalization. See [License](#license) for usage terms. ```python from lerobot.policies.flux3 import Flux3Policy from lerobot.policies.factory import make_pre_post_processors repo_id = "black-forest-labs/flux-3-action-so101" policy = Flux3Policy.from_pretrained(repo_id) preprocessor, postprocessor = make_pre_post_processors(policy.config, pretrained_path=repo_id) ``` Use the FLUX3 integration with shared-encoder Hub subfolder/revision support. Both frozen encoders load automatically from the pinned [base repository](https://huggingface.co/black-forest-labs/flux-3-action-base). This repository contains no duplicated encoder weights. ## Observation and action contract Two cameras: `observation.images.scene` and `observation.images.wrist`, in that order. Six state/action dimensions; joint-delta actions with absolute gripper. Eight observations, two visual snapshots and past-command conditioning. Predict 42 actions, execute 32 at 30 Hz, then replan. Saved processor pipelines and normalization state files are required and load from this repository. Sampling: four Euler steps, shift 6.93, guidance 3, seed 42, BF16. The checkpoint config and `dataset_statistics.json` contain the full contract. ## Task adaptation Use the included `lora.json` recipe with your dataset: ```sh hf download black-forest-labs/flux-3-action-so101 lora.json --local-dir models/so101 lerobot-train --config_path=models/so101/lora.json \ --policy.path=black-forest-labs/flux-3-action-so101 \ --dataset.repo_id=YOUR_DATASET --output_dir=outputs/so101-lora --job_name=so101-lora ``` The recipe inherits the robot/model settings from this checkpoint. It trains a rank-32 LoRA with bf16 mixed precision, batch size 2, gradient accumulation 4 and 10,000 steps. ## Training data The policy was trained on the SO-101 episodes of [lerobot/community_dataset_v3](https://huggingface.co/datasets/lerobot/community_dataset_v3). ## Documentation - [SO-101 with LeRobot](https://docs.bfl.ai/flux_3/flux3_action_so101) - [Run the model](https://docs.bfl.ai/flux_3/flux3_action_inference) - [Fine-tune](https://docs.bfl.ai/flux_3/flux3_action_finetuning) ## Out-of-scope use FLUX 3 Action outputs joint targets. Nothing in the model bounds joint velocity, force or workspace; the application must enforce those limits and keep a hardware stop within reach. Validate on a simulator or with the arm's safety limits engaged before running near people. The model and its derivatives may not be used: - In any way that violates applicable law. - To control a machine in a way that endangers people, without human oversight and a means to stop it. - For fully automated decision making or high risk applications that adversely impact an individual's legal rights or otherwise create or modify a binding, enforceable obligation. - To harass, abuse, threaten, stalk, or bully individuals or groups of individuals. - For the purpose of exploiting, harming or attempting to exploit or harm minors in any way. Nothing contained in this model card should be interpreted as or deemed a restriction or modification to the license the model is released under. ## Responsible AI development Black Forest Labs is committed to responsible model development and deployment. FLUX 3 Action outputs motor commands and, on request, predicted camera frames of the scene it is acting in. For information about our mitigations, evaluation processes and policies, see [Capable, Open, and Safe: Combating AI Misuse](https://bfl.ai/blog/capable-open-and-safe-combating-ai-misuse). To report safety concerns, contact [safety@blackforestlabs.ai](mailto:safety@blackforestlabs.ai). ## License This model falls under the [FLUX Kommunity License v.1.0](LICENSE.md). The text encoder in [flux-3-action-base](https://huggingface.co/black-forest-labs/flux-3-action-base) is an unmodified copy of Qwen3-VL-4B-Instruct under Apache-2.0. The code in [flux-action](https://github.com/black-forest-labs/flux-action) has its own license. ## Trademarks and IP This project may contain trademarks or logos for projects, products, or services. Use of Black Forest Labs and FLUX trademarks or logos in modified versions of this project must not cause confusion or imply sponsorship or endorsement. Any use of third-party trademarks, intellectual property or logos are subject to those third-party's policies.