Writing study: OmniForcing: Unleashing Real-time Joint Audio-Visual Generation
← ECCV 2026 oral writing corpus
Yaofeng Su, Yuming Li, Zeyue Xue, Jie Huang, Siming Fu, Haoran Li, Haoyang Huang, Nan Duan · Official ECCV 2026 oral record
Figure, table and section ledger
Editorial contribution type: Efficiency, inference & systems (primary).
Layout extraction, not a required paper template. Counts are distinct caption labels; a multi-panel figure counts once. Section spans are estimated from heading positions, including floats and whitespace. Method and limitations.
- Canonical PDF
- 19 physical pages
- Before references / appendix
- ≈ 15.15 page-equivalents
- Main figures / tables
- 4 / 4
- Attached appendix figures / tables
- 0 / 0
- Attached appendix span
- No boundary detected; not assumed zero
Extraction notes: No attached appendix boundary detected. Zero means no recognized caption, not independently verified absence.
Source-bound correction: All top-level main headings and the reference boundary checked against source PDF text; source-bound positions use physical single-column y/page-height, not a two-column reading-order surrogate. Running page-number/author titles, affiliations, numbered contribution bullets, footnotes and bibliography continuations are excluded. This linked manuscript contains no appended supplementary section; separate supplement links were not scanned.
Section-by-section lengths and source pages
The title/abstract/front matter precede the first numbered heading. A span ends at the next detected top-level heading, not at the last sentence; end-page whitespace is included. References and statements are listed separately. These are not measurements of text-only or artwork area.
All detected figure and table locations
Main text
Attached appendix
No recognized caption labels.
Argument and organization
- 1 Introduction; 2 Related Work
Frames bidirectional attention as a deployment obstacle for streaming rather than equating a smaller denoising step count with real-time operation.
- 3 OmniForcing: pipeline, mask, stabilization and joint self-forcing
Defines the target streaming factorization, explains modality-aligned masks, then covers the stages needed to stabilize causal distillation and inference.
- 4 Experiments; 5 Conclusion
Quality, synchronization, runtime, time-to-first-chunk and sustained FPS are not collapsed into a single speed claim; stabilization ablations report failed convergence explicitly.
Problem, claim and contribution
Frames bidirectional attention as a deployment obstacle for streaming rather than equating a smaller denoising step count with real-time operation.
Abstract and Section 1, pp.1–2 · Main paper
Argument and section order
Defines the target streaming factorization, explains modality-aligned masks, then covers the stages needed to stabilize causal distillation and inference.
Sections 3.1–3.4, pp.5–10 · Main paper
Figures, captions and visual explanation
The mask diagram links physical one-second macro-blocks to unequal audio/video token counts and separately displays all four attention directions with labeled query/key axes.
Figure 3, p.7 · Main paper
Tables, comparisons and evidence
Quality, synchronization, runtime, time-to-first-chunk and sustained FPS are not collapsed into a single speed claim; stabilization ablations report failed convergence explicitly.
Tables 1–3, pp.14–15 · Main paper
Appendix and reproducibility
The retrieved official PDF contains the main argument and references; separate supplementary material was not linked in the inspected event record, so no appendix content is counted as read.
Main PDF pp.1–19; official oral event 6046 · Main paper
Clarity and writing conventions
Describes stages with their separate jobs and places hardware, precision, warmup and repeated timing measurements near the streaming-decoder table.
Figure 2, p.6; Table 4, p.15 · Main paper
What to reuse—and what not to infer
Transferable lesson: Streaming papers need an attention/time diagram, startup latency and throughput with hardware, quality retention, and failed-stability settings.
Boundary: The reported approximate 35-fold wall-clock speedup, startup time and FPS have different denominators; they must not be substituted for one another.
Apply this lesson using the argument and evidence worksheet.
Inspection record and sources
- Checked
- 2026-10-02
- Reading status
- Writing-focused manuscript scan + selected visuals inspected
- Main text
- Writing-oriented scan across the manuscript main argument, physical PDF pp.1–15: framing and section roadmaps, method/argument explanations, figure and table captions, comparison protocols, conclusions and limitations. Selected full-page rasters were inspected. This is not exhaustive line-by-line technical reading, proof verification or experimental reproduction.
- Appendix / supplement
- Main-manuscript pointers to appendix/supplement and the reference boundary were checked. Separate supplementary PDFs were not retrieved or inspected; their contents are not included in inspected-page counts.
- Visually inspected PDF pages
- 7, 14
This is a writing-focused scan of the manuscript and the stated supplementary scope: framing, organization, visual explanation, evidence presentation and wording. It is not a line-by-line proof audit, reproduction of results or exhaustive scientific review.