{
  "docId": "019dd923-5e88-73ef-bd5d-06b04d219fea",
  "docSlug": "dd91c78f6570bf29",
  "documentTitle": "2023 Air Street Capital The State of AI Report 2023",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 50,
  "pageCount": 163,
  "prevPage": 49,
  "nextPage": 51,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "dense",
  "nDataPoints": 1,
  "notes": "The slide compares diffusion-based models vs. masked transformer models in the context of video generation.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "photo"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/50",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-50",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "MAGVIT is a masked generative video transformer... It currently has the best FVD on video generation benchmarks and it's 250x faster than video diffusion.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a226-ebdac3079aad",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.74,
        "x": 0.13,
        "y": 0.69
      },
      "kind": "image",
      "text": "Sequence of video frames showing a train in a fantasy landscape",
      "attrs": null,
      "subkind": "illustration",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "89686f06-a824-46e2-9802-84042ef07ff1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.94,
        "x": 0.03,
        "y": 0.21
      },
      "kind": "list",
      "text": "Similar to last year (Slide 33), the race is between video diffusion and masked transformer models (although algorithmically the two are very similar). Last year's Make-a-video and Imagen were based on diffusion while Phenaki was based on a bidirectional masked transformer.\n\nVideoLDM is a latent diffusion model capable of high-resolution video generation (up to 1280 x 2048!). They build on pre-trained image diffusion models to turn them into video generators by temporally fine-tuning with temporal alignment layers.\n\nMAGVIT is a masked generative video transformer. Similarly to Phenaki, it uses a 3D tokeniser to extract spatio-temporal tokens. It introduces a novel masking approach. It currently has the best FVD on video generation benchmarks and it's 250x faster than video diffusion.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "1aca9c39-daa4-4298-8c1b-37df3fdeba4d",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.46,
        "x": 0.27,
        "y": 0.87
      },
      "kind": "paragraph",
      "text": "\"The Orient Express driving through a fantasy landscape, animated oil on canvas\"",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f8da4618-aea0-41d5-802b-ec8d4edfc106",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.45,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "The text-to-video generation race continues",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e37dd3d4-9383-4ed4-a0e0-e2c2f1b32a35",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-41afd44ef59f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Research + Industry sections inventory model, compute, funding facts.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-523bfb7f96e6",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Escalating capabilities, compute concentration and capital flows.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 58,
      "from": 50,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-f70f6435b722",
      "evidence": "Text-to-video, instruction editing, Gaussian splatting, NeRFs, depth, SAM, DINOv2, weather, music.",
      "position": 6,
      "objective": "Generative media and visual foundation models proliferate",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 72,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}