ω-0, a latent predictive whole-body world-action model, enables real-world humanoid concurrent loco-manipulation by directly predicting controller-compatible whole-body action latents from language instructions, visual observations, and proprioceptive state. It learns compact future observation embeddings as a predictive objective, coupling latent visual foresight with diffusion-based action generation, and supports egocentric RGB, exocentric RGB, and exocentric depth inputs. The model uses controller-based simulation replay to ground human and public visual-motion priors into executable action latents. The authors collected ω-HOME, a 40+ hour real-world household humanoid dataset with synchronized multi-view observations, whole-body SMPL motions, robot states, and action latents. Real-world experiments on 11 household tasks show a single ω-0 model produces smooth manipulate-while-moving behaviors and outperforms representative imitation learning, VLA, humanoid, and world-action-model baselines, achieving 81.8% real-world success and 90.3% task progress with 1 unified policy. The framework uses a three-stage training pipeline: first, a FAST tokenizer converts continuous humanoid trajectories into discrete whole-body action tokens, and a pre-trained Qwen3-VL model learns action semantics; second, V-JEPA visual features, language, view tokens, and action-aware VLM features condition paired video and motion queries, with cross-attention injecting scene dynamics and an action DiT denoising SONIC-compatible action chunks;