SIGGRAPH Writing Style

A SIGGRAPH paper earns acceptance by making a graphics contribution legible on the first page and showing its results, not merely describing them. The reader is a domain expert who will form an opinion from your teaser figure and results video before finishing the abstract. This skill builds the SIGGRAPH first-page arc and the body discipline the acmart budget enforces. Anchor format facts to resources/official-source-map.md.

The SIGGRAPH first-page arc

problem a graphics practitioner recognizes -> why prior methods fall short (quality, speed, generality, or robustness) -> our technique in one sentence -> the result shown (teaser) -> what it enables. All of it visible on page one, with the teaser figure doing half the work.

• Lead with the visual/temporal problem: a rendering that is too slow or too noisy, a simulation that is unstable, a geometry operation that fails on real meshes. Not "deep learning has transformed graphics."
• State the contribution as a method plus a measurable gain — "Nx faster at equal quality," "converges where prior work diverges," "handles inputs prior methods cannot." A method with no quality/performance axis to move is not yet a SIGGRAPH contribution.
• The teaser must show the best result on a recognizable case and, ideally, a side-by-side with the strongest baseline. It is the single most-read object in the paper.

The teaser figure carries the paper

Spend disproportionate effort here:

• One glance should convey what problem, what input, what output, how much better.
• Prefer a real, hard scene over a toy; reviewers distrust teasers that only work on the easy case.
• If the contribution is temporal, the teaser points at the video ("see supplemental video") — but the still must still stand alone.

Body structure (conference/dual-track: <= 7 pages)

A workable graphics-paper skeleton:

1. Introduction — the first-page arc above; contributions as a short bulleted list.
2. Related work — positioned by what prior methods cannot do that you can (see siggraph-related-work), not a chronological survey.
3. Method — the technique, with the math and algorithm a reader needs to reimplement; push long derivations to the supplemental appendix.
4. Results — comparisons, ablations, timings, and quality metrics (see siggraph-experiments), each figure earning its space against the page budget.
5. Limitations — honest failure cases, shown not hidden; this builds credibility here.
6. Conclusion — one paragraph; no over-signposting.

Page-budget discipline (acmart)

• The conference-track body is <= 7 pages excluding references and up to two figures-only pages. Every figure competes with text for that space — cut a figure that does not change a reader's belief.
• Move derivations, parameter tables, and network architecture details to the supplemental appendix. The body carries the argument and the decisive results; the supplemental carries depth.
• Do not shrink fonts or edit the acmart class to recover space — that is a mechanical reject. Recover space by cutting, not by tampering.

Voice and honesty

• Show, then claim. "Our method removes the flicker (Fig. 4, video 0:30)" beats "our method is temporally coherent." Graphics reviewers believe pixels, not adjectives.
• Quantify every comparison with a metric and equal conditions; a qualitative "looks better" invites the reviewer to disagree.
• Name limitations before the reviewer does. A shown failure case with an explanation is far stronger than a concealed one the reviewer discovers in your video.
• Timings are claims — report them with hardware, resolution, and settings, or do not make them.

Anti-patterns

• A method paper with no comparison to the obvious prior work — the fastest path to reject.
• A teaser that only works on the easy input.
• Prose that describes results the figures/video do not actually show.
• Over-length body recovered by font/margin tampering.
• Limitations reduced to a boilerplate sentence, or omitted entirely.

Output format

[First page] graphics problem + inadequacy + technique + teaser + payoff all present? yes/no
[Teaser] shows best result on a recognizable case, ideally vs baseline? yes/no
[Contribution] method + measurable quality/performance gain stated? yes/no
[Body budget] conference-track <=7 pages, derivations moved to appendix? yes/no
[Honesty] comparisons quantified + limitations shown (not hidden)? yes/no
[Revision queue] <ordered>