Autoregressive video diffusion models have emerged as a foundation for long-horizon world rollout. However, when extended far beyond their training windows, these models suffer from progressive drift: visual quality deteriorates, motion becomes inconsistent, and object identity gradually collapses as generated frames are recursively reused as context. We study a supervision-horizon gap: short-horizon distillation leaves later rollout states without direct supervision. We show that matching short-window marginals need not determine long-horizon behavior, while longer on-policy trajectories expose delayed failures to training. This motivates temporal manifold matching at scale for long-horizon world models. Building on this insight, we introduce LongForcing, a distillation approach that scales teacher supervision to long student self-rollouts. LongForcing preserves the short-horizon teacher-forcing and causal ODE-distillation stages, while replacing the realscore teacher in the final distribution-matching stage with an extended-horizon bidirectional teacher. The long-horizon teacher supervises extended student self-rollouts using the standard distribution-matching formulation, without modifying the student architecture or increasing inference cost. We evaluate LongForcing on text-to-video generation and interactive action-conditioned world models, achieving a VBench Quality score of 85.52 and improving WorldRoamBench 60s action score by 23.9 points in first-person ablations. Qualitative stress tests further sample interactive rollouts up to 24 hours, far beyond the training horizon. These results suggest that scaling supervision along the temporal axis is a practical step toward long-horizon world models. The project website and code are at https://github.com/Fantasy-AMAP/LongForcing.
Figures & tables
Figure 1: LongForcing extends teacher guidance over student rollouts. Qualitative snapshots from the 5B interactive model span stress-test rollouts sampled up to 24 hours .
Figure 2: Visual and feature-space signatures of rollout drift. (a) Shared timestamps reveal late over-saturation and high-frequency artifacts. (b) Median absolute channel-mean shift from the early reference; dotted lines mark supervision horizons. (c) Median signed per-channel shift for the 81-frame model; persistent bands indicate a channel-wise direction shared across 20 videos.
Model
Size
WASD Action Interaction
Visual Quality
Action ↑
Ctrl. ↑
Partial Acc. ↑
Traj. ↑
Visual ↑
Aesthetic ↑
Aesth. Drift ↓
Imaging ↑
Proprietary Models
Genie 3 [ 2 ]
–
69.73
79.80
65.75
73.70
56.26
48.11
32.08
45.30
HappyOyster [ 46 ]
–
84.02
96.97
80.48
87.57
65.63
59.10
20.61
47.44
Open-Source Models
LingBot-World [ 47 ]
14B
18.81
23.23
18.50
19.12
15.99
14.32
81.70
13.84
Table 1: WorldRoamBench results. Except Ctrl., higher-is-better scores are multiplied by p=Ctrl./100 ; Aesth. Drift is 100[1−p(1−d)] for normalized drift d∈[0,1] . Best and second-best open-source results are bold and underlined.
Table 4
Frames
Approx. span
VBench (5s) Total ↑
VBench-Long (30s) ↑
Total
Quality
Semantic
Dynamic
81
∼ 5s
84.36
81.87
83.71
74.54
58.15
161
∼ 10s
83.76
78.02
81.59
63.77
56.67
241
∼ 15s
84.38
79.18
83.75
60.87
94.91
321
∼ 20s
84.56
82.83
83.99
78.19
60.93
Table 4: Supervision-horizon configurations for text-to-video generation. Horizons are rollout lengths in frames. Best and second-best scores are bold and underlined .
Frames
Approx. span
WASD Action Interaction
Visual Quality
Action ↑
Strict Acc. ↑
Partial Acc. ↑
Traj. ↑
Visual ↑
Aesthetic ↑
Aesth. Drift ↓
Imaging ↑
81
∼ 5s
64.39
42.89
63.01
87.27
45.05
39.81
56.58
46.40
161
∼ 10s
56.16
43.21
50.00
75.28
47.01
41.06
60.18
52.29
241
∼ 15s
83.39
78.24
85.14
86.78
55.19
49.93
43.30
50.25
321
∼ 20s
88.26
85.16
90.56
89.05
56.30
51.61
42.37
50.84
Table 5: Effect of the DMD supervision horizon on first-person interactive world modeling. The comparison varies the DMD supervision horizon with a fixed 1.3B backbone under the first-person WorldRoamBench 60s protocol; scores are unweighted. Higher is better except for Aesthetic Drift. The best and second-best results are shown in bold and underlined , respectively.
Figure 3: Rollout drift and human preference across supervision horizons. (a) Blinded pairwise preferences. (b) Mean Drift Score.
Figure 4: A–D: LongForcing rollouts over up to 54 minutes. E–F: LongForcing-5B (top) and LingBot-World v2 (14B; bottom) at matched timestamps from 0 to 599 seconds.
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
Figure A1: Mechanistic comparison of sparse local matching, inference-time anchoring, and LongForcing’s contiguous H -step on-policy trajectory matching under an extended-horizon teacher.
Model
#Params
Resolution
Throughput (FPS) ↑
Latency (s) ↓
VBench Total ↑
Quality ↑
Semantic ↑
Bidirectional video diffusion models
LTX-Video [ 52 ]
1.9B
768×512
8.98
13.5
79.83
81.88
71.62
Wan2.1 [ 43 ]
1.3B
832×480
0.78
103
83.37
84.30
79.65
Autoregressive video diffusion models
NOVA [ 53 ]
0.6B
768×480
0.88
4.1
80.31
80.66
78.92
Pyramid Flow [ 54 ]
2B
640×384
6.70
2.5
80.75
83.41
70.11
Appendix
Table 6: Full 5-second VBench comparison.
Model
Aggregate
Long-horizon consistency
Visual quality
Temporal evolution
Total
Quality
Semantic
Subject Cons.
Background Cons.
Aesthetic
Imaging
Dynamic
Temporal Style
LongLive [ 30 ]
82.80
83.21
81.17
98.36
96.25
63.77
68.35
38.89
24.15
Self Forcing [ 8 ]
81.49
82.67
76.77
97.76
95.00
61.72
67.70
41.93
23.56
Causal Forcing [ 10 ]
81.05
83.44
71.52
96.24
95.59
58.85
66.92
75.46
22.61
LongForcing (ours)
82.83
83.99
78.19
97.40
96.58
62.87
64.19
60.93
23.61
Appendix
Table 7: Full 30-second VBench-Long comparison.
Figure C1: Full time-resolved drift comparison. Curves show mean Drift Score with 95% bootstrap confidence intervals for text-to-video and interactive rollouts.
Figure D1: Qualitative 60-second text-to-video rollouts. Each row follows one LongForcing trajectory trained with 321-frame supervision; columns show frames at fixed video timestamps.
Figure D2: Qualitative comparison across supervision horizons. Upper and lower rows use 81- and 321-frame supervision, respectively, at matched rollout times from 0 to 59 seconds.