Stable-worldmodel is a research framework for goal-conditioned world models in reinforcement learning, providing a full pipeline from data collection and storage to model training and planning-based evaluation. The codebase is primarily a research platform with strong software engineering practices—pluggable data formats (Lance, HDF5, video, folder), a command-line interface (swm), and a Hydra-based configuration system for reproducible experiments. It includes implementations of multiple world model architectures (LeWM, PLDM, PreJEPA/DINO-WM, TD-MPC2, GCRL), planning solvers (CEM, MPPI, iCEM, gradient-based, Lagrangian), and environment wrappers with domain randomization across 30+ tasks (DeepMind Control, Gymnasium, PushT, OGBench, Craftax, robotics). The repository demonstrates high craftsmanship in its data engineering: the data format registry extends cleanly, the Lance format supports cloud-native streaming from Hugging Face Buckets, and utilities like GoalDataset, HistoryBuffer, and ReplayBuffer solve recurring RL data management problems. The project is actively maintained with extensive test coverage, comprehensive documentation, and a scalable training pipeline using stable-pretraining and Lightning. While the primary goal is research (testing hypotheses about world model generalization under variation), the codebase includes application-level production quality—CI/CD, CLI, checkpoint management, and robust error handling—making it a rare hybrid that serves both researchers and engineers.
Alpamayo 1 is a Vision-Language-Action (VLA) model designed for autonomous driving applications. It implements the Alpamayo-R1 architecture that combines a Qwen3-VL backbone for vision-language reasoning with a diffusion-based action expert for trajectory prediction. The system takes multi-camera video and egomotion history as inputs, generates Chain-of-Causation (CoC) reasoning traces, and outputs future trajectories (6.4s horizon at 10Hz). Key components include a unicycle kinematic model for action space conversion, CUDA graph acceleration for efficient repeated inference, and robust geometric transformations. While the repository is noted to receive only limited maintenance as newer versions exist, the current implementation demonstrates strong engineering practices with several reusable components.
This repository provides the data pipeline for the Sekai dataset, a large-scale egocentric video dataset for world exploration and generation research. It contains two main toolchains: dataset_downloading (extracting YouTube URLs from CSV annotations and downloading videos via yt-dlp) and clip_extracting (a multi-step GPU-accelerated pipeline to split raw videos into annotated clips, transcode them to H.265, and remux with audio). The pipeline is designed for the Sekai-Real subset and requires NVIDIA GPUs with NVENC support, CUDA 11.8, and specific Python dependencies. The code is well-structured with clear separation of steps, logging, and multiprocessing support. While tightly coupled to the Sekai dataset and hardware requirements, the utility modules for GPU video decoding/encoding and frame sampling are reusable in other video processing projects.
This is a research-grade PyTorch implementation of Danijar Hafner's Dreamer v4 model-based reinforcement learning algorithm. The core artifact is the `dreamer4` package, dominated by a monolithic `dreamer4/dreamer4.py` (~3000 lines) that implements `VideoTokenizer`/`SpaceTimeTokenizer`, `DynamicsWorldModel`, `AudioTokenizer`, and `StateTokenizer`. These classes build causal space-time transformers with separate space/time attention blocks, latent flow-matching diffusion with shortcut consistency, BYOL-style tokenizer pre-training, PoPE attention, and a large zoo of experimental loss terms (latent autoregressive loss, decorrelation, orthogonality, SigReg, self-flow distillation, shortcut loss). `dreamer4/trainers.py` provides training loops (VideoTokenizerTrainer, BehaviorCloneTrainer, DreamTrainer, SimTrainer) and `dreamer4/cli.py` exposes CLI entry points for training and serving the world model. The repository is hybrid: primarily research code, but packaged and tested like a library. It ships a pyproject.toml with pinned dependencies, a parametrized test suite (tests/test_dreamer.py, test_e2e.py, test_env_wrapper.py), a browser-based Snake environment with a Flask-style server (`dreamer4/web_env/`) used for human demonstration collection and replay-buffer inspection, and environment wrappers in `dreamer4/env.py` that record agent experience to folders or memmap replay buffers. Root-level scripts are the actual experiment entry points, targeting standard benchmarks: HalfCheetah (imagination RL), CartPole (dynamics + RL), MovingMNIST (tokenizer + dynamics), and Snake (PPO with dynamics model). Craftsmanship is above typical research code: clean package layout, pytest coverage across many hyperparameter combinations, self-contained utilities (LossNormalizer, tensor pack/unpack helpers, with_seed decorator, memmap replay-buffer helpers), and HuggingFace Accelerate integration for distributed/mixed-precision training. However, the single-file core, 100+ hyperparameter flags on trainers, scattered hard-coded constants, sparse docstrings, and reliance on volatile niche packages (x-transformers, h-net-dynamic-chunking, PoPE_pytorch, adam-atan2-pytorch) make long-term stability and code reuse risky. The repository is recently committed and appears maintained, but it is explicitly beta-status experimental research code, not a production system.
This repository is the official implementation of the SANA family of diffusion models for high-resolution image and video generation, developed by NVIDIA Research. It is primarily a research codebase, containing training and inference pipelines for multiple model variants: SANA (text-to-image), SANA-1.5, SANA-Sprint (few-step distillation), SANA-Video, SANA-Video 2.0, SANA-Streaming (long video), LongSANA, and SANA-WM (world model for camera-controlled video). The codebase also includes post-training with reinforcement learning (Sol-RL), ControlNet support, and various schedulers (flow matching, DPM-Solver, LCM, self-forcing). While the core purpose is research, the repository also provides deployable Gradio applications, conversion scripts to Hugging Face diffusers format, and integration with ComfyUI and SGLang, giving it a hybrid application character. The code is heavily optimized with custom Triton kernels for efficient attention (LiteMLA, Gated Delta Net) and includes distributed training utilities (FSDP, context parallelism). The repository is actively maintained with frequent commits and supports a wide range of model sizes (600M to 14B parameters).
Helios is a real-time long video generation system that implements text-to-video (T2V), image-to-video (I2V), and video-to-video (V2V) generation using a 3D transformer architecture. The system features a modular design with specialized optimizations for performance including Ahead-of-Time Compilation (AOTI), group offloading for VRAM reduction, context parallelism (Ulysses/Ring attention), and Flash Attention kernel optimizations. It employs a multi-stage training pipeline (Base → Mid → Distilled) to balance quality and sampling speed. The repository includes a Gradio web interface (app.py) for interactive generation, command-line inference scripts, training utilities, and a comprehensive evaluation suite measuring video quality across aesthetic, motion, semantic, and naturalness dimensions. While containing research elements for ablation studies and benchmarking, its primary function is as a deployable video generation application.
This repository is a curated literature review (awesome list) focused on the progression from video generation to world models in AI research. It systematically organizes and categorizes recent academic papers (primarily from 2022-2026) into a hierarchical structure covering technical foundations, application domains, evaluation methodologies, and theoretical perspectives. The main value lies in its comprehensive paper collection with structured metadata including publication dates, venues, paper links, project pages, and official code repositories. The repository follows a standard awesome-list format using markdown tables with shields.io badges for easy navigation to paper resources. Content is organized into eight main sections: Generation 1-4 (covering faithfulness, interactiveness, planning, and randomness aspects), Evaluation and Datasets, Study and Rethinking, Downstream Tasks, and World Models for Other Applications. Each section contains subsections detailing specific technical approaches (e.g., GAN-based, U-Net-based, DiT-based video generation) and application areas (robotics, autonomous driving, gaming). The list is actively maintained with frequent updates incorporating very recent pre-print papers. As a static reference resource, the repository contains no executable code, data processing pipelines, or computational components. Its reproducibility is high because the curated content is directly available in the README upon cloning, requiring no setup or dependencies to access the paper list. The organizational pattern demonstrates good practices for academic literature curation, though it does not contain novel reusable data engineering components beyond the standard markdown documentation format.
This repository implements Geometry Forcing, a research framework for video generation with geometric consistency. The core contribution marries video diffusion models (DFoT - Diffusion Forcing Transformer) with 3D representation learning (VGGT) to produce temporally consistent videos conditioned on camera poses. It includes full training pipelines (Hydra + PyTorch Lightning), a suite of video metrics (FVD, FID, LPIPS, VBench), and evaluation scripts for reprojection/revisit error using DROID-SLAM. The codebase is structured as a research template with reusable base classes for algorithms, data modules, experiments, and metric registries. Raw datasets (RealEstate10K, Kinetics-600, Minecraft) are downloaded and preprocessed into video clips with associated camera parameters. Transformations include VAE encoding for latent diffusion, augmentation (flip, reverse), and sliding-window clip extraction. The primary research question is whether video diffusion can be constrained by explicit 3D geometry to reduce inconsistencies in generated videos, measured by geometry-aware metrics. The repo also demonstrates composable sampling via History Guidance and an alignment loss that projects diffusion latents into VGGT feature space. While the code is well-engineered for research (modular design, config-driven, metric registry), it carries typical research code baggage: hardcoded paths, vendored third-party code (DROID-SLAM, VGGT), potential dependency pinning conflicts, and some dead/commented code. Reproducibility is moderate: dependencies are pinned but no environment definitions beyond requirements.txt, and dataset downloads rely on fragile YouTube scraping.