Speculative Denoise–Verify Video Diffusion Diagram (Editable PowerPoint)
Overview
This editable PowerPoint diagram shows speculative, verifier-gated denoising for autoregressive video diffusion. It is adapted from the original paper figure under CC BY 4.0.
What the diagram shows
Row (a) shows the baseline, where every token grid runs the full Denoise chain. Row (b) alternates Denoise and Verify steps, so converged tokens are accepted early. A zoom-in panel walks through k-step denoising, a draft, a diamond verifier and an Accept/Reject decision based on a predicted-convergence score, with a green and orange token legend. Every box, token cell, check mark, arrow and label is a native shape you can recolor, resize and relabel for your own slides, posters or papers.
File format
The download is a single-slide .pptx file measuring 10.23 × 5.02 in. It opens in PowerPoint, Keynote, WPS and LibreOffice Impress. Google Slides works too, although small layout shifts are possible. A full-resolution preview.png and a 1200 px web preview are included.
Shapes and editing
The figure is built from 273 native editable shapes: 164 auto-shapes (token cells, Denoise and Verify blocks, panels and the diamond), 50 text boxes, 36 freeforms (the check marks) and 23 lines or connectors. It contains no embedded images, so everything is vector. You can edit the token states (converged, accepted or rejected), block colors, arrows, the dashed zoom frame, the Accept/Reject boxes and every label.
Math labels and fonts
The math labels (x̂k0, relk(i), σt, ε) are built from small positioned text boxes and sub/superscript runs, not PowerPoint equation objects. If you retype them, you may need to nudge the hat or the scripts. The text uses Calibri for labels and Times New Roman for math; LibreOffice substitutes the metric-compatible Carlito and Liberation Serif. This is a faithful redraw for reuse, not the authors' original source file.
Source & license Adapted from Figure 3 of Speculative Forcing for Efficient Autoregressive Video Generation by Hyunwook Choi, Seungryong Kim, Dongyoon Han, Sangdoo Yun (arXiv:2610.22347, https://arxiv.org/abs/2610.22347; code: https://github.com/naver-ai/speculative-forcing). The original is licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Converted to editable PowerPoint shapes; the panel (b) title was made generic. Changes made by 叫我Alonzo就好了/alonzolab. Not endorsed by the original authors. Please keep the attribution (see LICENSE-ATTRIBUTION.txt) when you reuse the figure.