Video World Models · 2026-10-01
FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation
Bo Yin, Xiaobin Hu, Jiaqi Zhao, Shuicheng Yan
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
FrameMorrow selects historical frames for long-horizon video generation by predicting prospective tokens that represent future information needs. The selector outputs explicit historical frames (not internal states), enabling plug-and-play use across diverse generators, including closed-source models, with little additional inference cost. Across five benchmarks and 11 generative models, the authors report improved long-range consistency, visual quality, and action alignment.
Why it matters
Long-horizon video generation becomes expensive and redundant as history grows, and past content that matches the present may not help future generation. FrameMorrow changes selection from current-context relevance to future-information-need relevance, and outputs explicit frames for compatibility across generators. This supports better long-range coherence and (in world models) action alignment under a fixed historical reference budget.
Method
- FrameMorrow predicts a small set of prospective tokens from eligible history, recent context, and the known rollout condition, without generating the full future.
- The prospective tokens score eligible historical frames via attention; the method selects explicit historical frames (top-K) ordered by timestamp for the backbone.
- Training uses future-grounded ranking distillation from realized continuation using a frozen DINOv2 encoder, while inference uses only observed inputs.
Limitation
The paper states that the future-grounded teacher correspondence target is a visual relevance proxy: it “does not imply downstream benefit” and the proxy is not a measurement of incremental generation utility beyond the recent context or a guarantee of coverage across the continuation.
Abstract (from arXiv)
Long-horizon video generation requires models to effectively leverage an increasingly long generation history. As the generated history grows, retaining all previous content becomes increasingly expensive and redundant, making effective historical selection essential. Existing approaches often determine historical relevance based on the current content. However, information relevant to the present is not necessarily useful for future generation, while seemingly less relevant history may become important later. Our key insight is that historical information should be selected according to its relevance to future information needs. Capturing these needs does not require generating the full future; instead, a compact representation of what becomes important next is sufficient to guide historical selection. Building on this insight, we propose FrameMorrow, a prospective frame selector that predicts a small set of prospective tokens representing future information needs and uses them to identify relevant information from history. FrameMorrow selects explicit historical frames rather than model-specific internal states, enabling plug-and-play integration across diverse generators, including closed-source models, with little additional inference cost. We evaluate FrameMorrow across five benchmarks and 11 generative models spanning long-video generation, interactive generation, and action-conditioned world models. Extensive experiments demonstrate consistent improvements in long-range consistency, visual quality, and action alignment across diverse generation settings.
Related papers
- FutureWorld: A Plug-and-Play Causal Next-State Predictor for Interactive Video World Exploration
- History Forcing: Distilling History-Dependent Continuations for Long-Horizon Video World Models
- LongTake: Learning to Sustain Dynamics in Long-Horizon Video Generation
- GROME: GROunding Factorized MEmory into Video World Models
- Reading, Writing, Sharing, and Interacting with the World through Video
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-09-30 · Indexed by Pub-AI: 2026-10-01 UTC
FrameMorrow: Future-guided Frame Selection with Prospective Tokens for Long-Horizon Video Generation
Preprint posted: 2026-09-30 · Original record ↗
- arxiv
- 2609.38839
Explicit source versions (1)
- v1 · 2026-09-30
- Preprint available
Source date: 2026-09-30 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Long-horizon video generation requires models to effectively leverage an increasingly long generation history. As the generated history grows, retaining all previous content becomes increasingly expensive and redundant, making effective historical selection essential. Existing approaches often determine historical relevance based on the current content. However, information relevant to the present is not necessarily useful for future generation, while seemingly less relevant history may become important later. Our key insight is that historical information should be selected according to its relevance to future information needs. Capturing these needs does not require generating the full future; instead, a compact representation of what becomes important next is sufficient to guide historical selection. Building on this insight, we propose FrameMorrow, a prospective frame selector that predicts a small set of prospective tokens representing future information needs and uses them to identify relevant information from history. FrameMorrow selects explicit historical frames rather than model-specific internal states, enabling plug-and-play integration across diverse generators, including closed-source models, with little additional inference cost. We evaluate FrameMorrow across five benchmarks and 11 generative models spanning long-video generation, interactive generation, and action-conditioned world models. Extensive experiments demonstrate consistent improvements in long-range consistency, visual quality, and action alignment across diverse generation settings.
