Adaptive Tokenisation Via Temporal Redundancy Masking And Latent Inpainting

Adaptive video tokenisation seeks to dynamically allocate token budgets based on the underlying visual complexity of a sequence. Current continuous-regime approaches achieve this via iterative binarised searches or trained neural regressors, while discrete methods often require a full-rate decoder pass to estimate information content. We demonstrate that such computational overheads are not strictly necessary. We show that the latent space of a frozen continuous video tokeniser inherently encodes temporal redundancy that can be exploited directly: spatial positions whose latent representations change minimally between consecutive frames carry near-zero additional information. We introduce a parameter-free adaptive token allocation mechanism that applies a fixed threshold to per-position temporal-L1 differences, identifying and dropping redundant latent positions. Consequently, the compression rate emerges naturally from the input content rather than being enforced top-down: static scenes get compressed aggressively, while highly dynamic sequences retain more tokens. To reconstruct the dropped positions, we propose the Latent Inpainting Transformer (LIT), a lightweight factorised spatial-temporal attention architecture. The resulting inference pipeline is highly efficient, requiring only a single encoder pass and one LIT forward pass, eliminating the need for auxiliary routing networks. Evaluations across TokenBench and DAVIS, which are the standard benchmarks used by recent tokenisers~\cite{infotok, agarwal2025cosmos}, indicate that our framework yields meaningful, content-driven token allocation while maintaining competitive reconstruction fidelity, and delivers a $31\times$ inference-time speedup over the continuous adaptive baseline (ElasticTok-CV) and an $\approx2\times$ speedup over the discrete information-theoretic baseline (InfoTok)

Audio-visual training for improved grounding in video-text LLMs

Recent advances in multimodal LLMs, have led to several video-text models being proposed for critical video-related tasks. However, most of the previous works support visual input only, essentially muting the audio signal in the video. Few models that support both audio and visual input, are not explicitly trained on audio data. Hence, the effect of audio towards video understanding is largely unexplored. To this end, we propose a model architecture that handles audio-visual inputs explicitly. We train our model with both audio and visual data from a video instruction-tuning dataset. Comparison with vision-only baselines, and other audio-visual models showcase that training on audio data indeed leads to improved grounding of responses. For better evaluation of audio-visual models, we also release a human-annotated benchmark dataset, with audio-aware question-answer pairs.