AI Video2026
Provides a DiT-based audio–video foundation model plus an official Python inference and LoRA trainer. Ships multiple production-ready pipelines (text/image/audio→video), checkpoints, and performance optimizations (FP8, distilled pipelines) for high-fidelity synchronized audio–video generation.
AI Audio2026
Real‑time full‑duplex speech‑to‑speech system that controls conversational role via text prompts and voice timbre via audio-conditioned embeddings. Built on Moshi; optimized for low-latency, persona-consistent spoken interactions.
AI Audio2026
Generates low-latency, streaming text-to-speech entirely on CPUs (no GPU or cloud API required), using an ~100M-parameter model with voice cloning and multilingual support. Optimized for low resource use (2 CPU cores, ~200ms to first audio chunk) — suited for local, privacy-sensitive, or embedded TTS.
Provides a conditional memory module that performs O(1) N‑gram lookups and fuses static embeddings into transformer hidden states — enables offloading large embedding tables to host memory with minimal inference overhead.
AI Audio2026
Local-first voice cloning studio that runs on your machine to clone voices, generate speech in 23 languages, apply audio effects, and compose multi-voice projects. Includes five switchable TTS engines, a REST API, and native GPU/MLX support for privacy-sensitive offline workflows.
Embodied AI2026
Enables research-grade character animation with neural networks in a single NumPy/PyTorch environment — train models, run inference, and visualize results without leaving Python. Includes ECS-style architecture, mocap import (GLB/FBX/BVH), built-in renderer, and headless/standalone modes for rapid prototyping.
AI Audio2026
Generates high‑fidelity, expressive speech and environmental sounds from text. The MOSS‑TTS Family provides specialized models for long‑form TTS, multi‑speaker dialogue, voice design and realtime streaming, plus torch‑free inference paths (llama.cpp / ONNX) and Hugging Face releases.
AI Train2026
A challenge repository for training the best language model that fits inside a 16,000,000‑byte (16MB) submission artifact; provides baseline training code, FineWeb bpb evaluation, a public leaderboard, and compute-grant instructions for short 8×H100 runs.
AI Video2026
Unmixes green‑screen pixels with a neural model to recover straight (unmultiplied) foreground color and a clean linear alpha for every pixel, preserving hair, motion blur and translucency. Produces VFX‑standard EXR outputs, supports optional AlphaHint generators (GVM/VideoMaMa) and Docker/consumer‑GPU optimizations.
AI Agent2026
Lets an LLM autonomously propose, edit, run, and evaluate short single‑GPU LLM training experiments — fixed 5‑minute runs (~12 experiments/hour). Agent edits a single train.py; humans supply goals via program.md. Single‑GPU, val_bpb metric.
AI Model2026
Generate text, images, video, audio and action/robot trajectories from combined text, image, video, audio and action inputs. A Mixture-of-Transformers omnimodal foundation model (Cosmos3‑Nano, 16B params) focused on Physical AI (robotics, AV, simulation) and optimized for NVIDIA GPU runtimes.
AI Model2026
Instruction-tuned Mixture-of-Experts multimodal model that generates text from text+image inputs while activating a 4B subset of parameters for faster inference; supports a 256K context window, multilingual vision-language tasks, and is available under Apache-2.0.