\n<\/p>\n
Video generation from text has come a long way, but it still hits a wall when it comes to producing longer, multi-scene stories. While diffusion models like Sora<\/a>, Veo<\/a>, and Movie Gen have raised the bar in visual quality, they\u2019re typically limited to clips under 20 seconds. The real challenge? Context. Generating a one-minute, story-driven video from a paragraph of text requires models to process hundreds of thousands of tokens while maintaining narrative and visual coherence. That\u2019s where this new research from NVIDIA, Stanford, UC Berkeley, and others steps in, introducing a technique called Test-Time Training (TTT) to push past current limitations.<\/p>\n Transformers, particularly those used in video generation, rely on self-attention mechanisms<\/a>. These scale poorly with sequence length due to their quadratic computational cost. Attempting to generate a full minute of high-resolution video with dynamic scenes and consistent characters means juggling over 300,000 tokens of information. That makes the model inefficient and often incoherent over long stretches.<\/p>\n Some teams have tried to circumvent this by using recurrent neural networks (RNNs<\/a>) like Mamba or DeltaNet, which offer linear-time context handling. However, these models compress context into a fixed-size hidden state, which limits expressiveness. It\u2019s like trying to squeeze an entire movie into a postcard, some details just won\u2019t fit.<\/p>\nWhat\u2019s the Problem with Long Videos?<\/h2>\n
How Does TTT (Test-Time Training) Solve the Issue?<\/h2>\n