Provides ComfyUI-compatible conversions and LoRA adapters of the MiniMax‑H3 video+audio generative model, with example presets and demo videos to run short stereo audio+video inference inside ComfyUI workflows.
Provides raw, unscripted first-person household video footage for training vision and embodied AI models. Released incrementally on Hugging Face in WebDataset shards with metadata parquets under Apache‑2.0; current raw tier contains ~7,834 hours (≈397k videos).
Provides 90,000 hours of head-mounted egocentric video paired with synchronized 3D hand pose and an optional 3D full‑body pose add-on, with event-level semantic labels available as a complimentary layer — designed for embodied AI and robotics training at scale.
Provides 1,000 five-second video clips generated by MiniMax H3 for lightweight evaluation of multimodal generation and understanding. Clips are roughly 768p base resolution with diverse aspect ratios and themes, produced with a pruned int8 minimax_h3_fl2va checkpoint at 30 steps.
A LoRA adapter for MiniMax H3 that improves photorealistic rendering of people—preserving skin texture, coherent micro-expressions, film-style lighting and subtle handheld motion. Trigger word: r34l1sm; intended for text-to-video portrait and close-up shots.
Encodes videos into a Film Knowledge Graph and reconstructs them to learn agent-native, editable video representations for agentic reasoning and manipulation. Uses agentic auto-encoding with dual-loop textual-gradient optimization, reports large reconstruction gains, and releases a benchmark and dataset.
Builds an editable, persistent 3D world state to drive iterative previsualization for film, games, and design — enabling local edits and recombinations instead of one-shot video regeneration. Uses separate stages for state construction, state evolution, and state access, with render-feedback camera refinement.
Externalizes persistent scene state into a camera-indexed world bank and designs a long-horizon teacher whose sparse-attention supervision is distilled into a three-step student, enabling responsive, low-latency interactive long-horizon video generation with bounded denoiser context.
Predicts future video frames conditioned on an observed frame, a language instruction, and a sequence of end-effector poses and gripper states for robot manipulation. Uses per-arm SE(3) geometric encoding (PRoPE-style), a lightweight depth branch, SAM3 masks with a frozen V-JEPA teacher, and distribution-matching distillation for efficient, consistent action-conditioned rollouts.
Provides a 27B Qwen3.8 GGUF build for local/offline deployment, optimized with Unsloth Dynamic V3.0 quantization. Offers switchable thinking-mode, native vision-language understanding, and native long-context support (262k+ tokens).
A 27B Qwen3.8 vision‑language causal transformer quantized to NVFP4 for lower‑memory inference. Provides 262K native context (extensible to 1M), Unsloth Dynamic V3.0 4‑bit quantization and MTP support so Qwen3.8‑class multimodal workloads can run on 24GB‑class GPUs.
Provides a large-scale benchmark and a human-aligned metric for humanoid whole-body motion tracking — about 153 hours of optical mocap from professional performers plus HumanScore trained on 12K human-labeled preference pairs to reveal contact, timing, and stability failures.