Game World Model NPC Role-Conditioning Diagram

Overview

A three-panel diagram of a video-diffusion game world model that steers a player and an NPC separately. It is an editable PowerPoint diagram adapted from the original paper figure under CC BY 4.0.

What the diagram shows

Left: spatial role binding — the first frame X₀, per-role masks m₁, m₂ multiplied by handle embeddings h₁, h₂, the binding map B = Σ_i m_i h_i and a first-frame offset αB, plus a latent-time causal attention mask over z₀ … z_t (visible history vs. future). Centre: video latents → patch embedding → a DiT block (causal self-attention → U → role-conditioning layer → FFN) → predicted video. Right: role-specific control prompts ("Throw a fireball" for the player, "If a fireball approaches, jump to evade." for the NPC) → a frozen text encoder with projection → control tokens E₁, E₂ with key-only handles h₁, h₂, a background context (s₀, h₀), and cross-attention with Q from the video tokens U and K, V from the role tokens, added back as updated video tokens. Every box, arrow, label and math term is a native shape you can recolor, resize and relabel for your own slides, posters or papers.

File format

The download contains .pptx (1 slide, 13.33 × 7.75 in; opens in PowerPoint, Keynote, WPS and LibreOffice Impress; Google Slides works too, but complex freeforms may shift slightly), .svg and .drawio. Includes preview.png (full resolution) and a 1200 px web preview.

Shapes and editing

The file has 442 native editable shapes: 276 auto-shapes, 51 text boxes, 12 freeforms, 103 lines/connectors (2 small groups). No embedded images. Everything is vector. You can edit all text and math (X₀, m_i, h_i, B = Σ_i m_i h_i, αB, z₀ … z_t, E₁, E₂, (s₀, h₀), U, Q, K, V), every cell of the mask grids, binding map and causal attention mask, the stacked latent grids, DiT block, prompt cards with native vector role icons, the 3D control-token bars, handle badges, ⊕ nodes, dashed zoom links and all connectors.

Embedded images and fonts

This file has no embedded images; everything is vector. The first-frame and predicted-video thumbnails are simplified vector placeholder scenes (flat shapes, not screenshots); paste your own frames over them if you need real footage. Body text uses Calibri and formulas use Times New Roman. Calibri ships with Microsoft Office and Times New Roman is preinstalled on Windows and macOS (LibreOffice substitutes the metric-compatible Carlito and Liberation Serif). Sub- and superscripts are real formatted text runs, not equation objects. The SVG keeps text as text; the draw.io file is a shape-by-shape conversion and may need small nudges. This is a faithful redraw for reuse, not the authors' original source file.