ANTI-AI ARCHIVE
ART-HISTORY NODE / 331 · 2025-06-09

Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Huang, Xun; Li, Zhengqi; He, Guande; Zhou, Mingyuan; Shechtman, Eli

Introduction

Self Forcing trains video models on their own generated context, addressing the mismatch between ground-truth training frames and self-generated inference history in streaming generation.

ORIGINAL DOCUMENT#331
Figure 1: training-context comparisons for Teacher Forcing, Diffusion Forcing and Self Forcing.View full image ↗

Figure 1: training-context comparisons for Teacher Forcing, Diffusion Forcing and Self Forcing. Source: v2; the image version is distinct from the initial submission date.

Huang, Xun; Li, Zhengqi; He, Guande; Zhou, Mingyuan; Shechtman, Eli · original paper / arXiv · Rights in the paper and depicted works remain with their holders. Research quotation does not establish an open licence; republication rights await independent review.

Source · Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion ↗

RESEARCH ACCOUNT

Self Forcing trains video models on their own generated context, addressing the mismatch between ground-truth training frames and self-generated inference history in streaming generation.

ANTI-AI ARCHIVE · Revised 2026-10-03

What the paper investigates

Training performs autoregressive rollout with KV caching and a video-level objective. Few-step diffusion and gradient truncation control computation, while a rolling cache supports continuation. Reported streaming speed is conditional on the tested hardware and configuration.

Section sources: Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Its place in generative-art history

Editorial interpretation: output can arrive over time instead of only after a complete clip is prepared. This supplies a different condition for interactive imagery, while a streaming research demonstration remains distinct from an actual artistic practice.

Section sources: Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Reading limits and versions

Low-latency demonstrations do not guarantee stability across devices, prompts and durations. The account reads the revision and sorts by initial submission.

Section sources: Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Sources for this account

Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion ↗

Read the arXiv abstract, authors and version history, plus the selected figure/page on PDF page 2; proofs and full experiments were not independently audited.

Read the arXiv abstract, authors and version history, plus the selected figure/page on PDF page 2; proofs and full experiments were not independently audited. The account and translation are AI-assisted, pending independent human review. Section references identify evidence without claiming independent verification of every historical statement.

Continue with a comparative question

Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models ↗

Self Forcing: compare methods, control or evaluation conditions with FramePack.

HunyuanVideo ↗

Self Forcing: compare methods, control or evaluation conditions with HunyuanVideo.

Realtime Video ↗

Self Forcing: compare methods, control or evaluation conditions with Realtime Video.

Dates & version record

Date displayed for this node: 2025-06-09 · Historical date recorded for the source: 2025-06-09

The timeline uses the initial arXiv submission, distinct from conference publication, model release and the archive addition on 3 October 2026. The account and image refer to v2; its date is recorded in the source history.

These dates refer to the historical event or recorded version, not this page’s publication date. The original date precision and unresolved questions are retained.

Original sources & further reading

These links lead to the cited paper, article, institution or conference page. External texts retain their source languages.

01
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion ↗

Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

https://arxiv.org/abs/2506.08009

Current cited URL

Read the arXiv abstract, authors and version history, plus the selected figure/page on PDF page 2; proofs and full experiments were not independently audited.

Provenance, translation & verification

Archive node #331 · Initial research-paper submission; methods, evaluation and production conditions

Research materials & supplement references · 1

Recent research papers: additions and reconciled sources since 2022 · Read the arXiv abstract, authors and version history, plus the selected figure/page on PDF page 2; proofs and full experiments were not independently audited. · 033
Date as recorded: 2025-06-09
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion ↗
Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

Archive account based on the listed research materials, not a full translation of the linked work. AI-assisted translation; independent human review pending.

Official material was read within the stated scope; see the source note for reading limits and outstanding checks.

Cite this node

ANTI-AI ARCHIVE. Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion. Art-history node #331. https://salondesrefuses.cn/en/art-history/356

For specific historical claims, also cite the original sources above and include your access date. This account is not a full translation of the linked work.

Adjacent nodes follow chronological order; adjacency does not establish direct influence or causation.