README.md

metadata

title: RxBrain Embodied Cognition
emoji: 🔮
colorFrom: red
colorTo: yellow
sdk: gradio
sdk_version: 5.42.0
app_file: app.py
short_description: Embodied reasoning + visual imagination (Hy-RxBrain 1.0)
python_version: '3.12'
startup_duration_timeout: 1h

RxBrain — Embodied Cognition Foundation Model

Interactive demo of tencent/Hy-Embodied-RxBrain-1.0, a unified Mixture-of-Transformers model that couples language reasoning with visual imagination.

Three capabilities:

• Visual QA — question answering over images.
• Text-to-Image — imagine an image via the flow-matching head (decoded through a frozen FLUX VAE).
• Embodied Planning — decompose a task into interleaved steps, emitting both the action (text) and the imagined goal frame (image) per step.

Runs on ZeroGPU. Imagination requires the external FLUX VAE (ae.safetensors), fetched from black-forest-labs/FLUX.1-schnell.