{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreidshk5bpngbjjs5tpjfs5ymitn5cj6awzxp3cgyd4xdgacvpejvqi",
    "uri": "at://did:plc:lk3jfj3zq4k4wxnk474axylu/app.bsky.feed.post/3momn5eofl4q2"
  },
  "path": "/t/hinoko-space-method-v0-0-2/1384098#post_1",
  "publishedAt": "2026-06-19T05:24:06.000Z",
  "site": "https://community.openai.com",
  "textContent": "Based on testing over the past few days, I have updated the Hinoko Space Method.\n\nThe Method now includes some new explanations and more methods.\n\nHinoko Space Method\n\n## Methods\n\n### 1. Scene Construction\n\nScene construction begins by establishing a reusable three-view spatial structure derived from either real-world references or AI-assisted spatial layouts.\n\nThe three-view structure serves as a spatial anchor, supporting environmental continuity and preserving scene consistency across generated outputs.\n\n#### 1-1 Why are Three-View References Necessary?\n\nAI-generated scenes may exhibit the following issues:\n\n  * Spatial drift\n  * Environmental detail inconsistency\n  * Perspective instability\n\n\n\nBy providing multiple spatial viewpoints, creators constrain spatial interpretation while allowing AI systems to infer and complete missing visual information.\n\nThree-view references are not a new invention.\n\nThey are a long-established visualization method used in design, architecture, engineering, and concept art.\n\nHinoko-Space-Method does not claim to invent three-view references.\n\nInstead, it applies multi-view spatial references to AI-assisted narrative generation workflows in order to reduce spatial drift and improve continuity.\n\n#### 1-2 Core Principles\n\nThe objective is not to delegate world creation to AI, but to extend human creative capability through structured spatial guidance.\n\nThe creator remains responsible for defining the world, while AI functions as an assistive tool for visualization and scene expansion.\n\n* * *\n\n### 2. Character Construction\n\nCharacter construction utilizes one or more reference images to preserve character identity and reduce visual distortion across generated scenes.\n\nThe objective is to maintain consistency in appearance while allowing flexible scene adaptation.\n\n#### 2-1 Usage Principles\n\nReference materials may include:\n\n  * Face references\n  * Full-body references\n  * Clothing references\n  * Expression references\n\n\n\nWhen using real-life references, appropriate authorization and consent should be obtained before generation.\n\n* * *\n\n### 3. Combined Use\n\nBy combining three-view spatial references, character references, and designated audio direction, creators can guide AI-assisted video generation while maintaining narrative and visual consistency.\n\nThis workflow enables:\n\n  * Improved scene consistency\n  * Improved character consistency\n  * Reduced visual drift\n  * More controllable video generation\n\n\n\nThe purpose is not to replace creative decision-making, but to provide a structured framework for creator-guided generation.\n\n* * *\n\n### 4. Temporal Continuity\n\nWhen generating a subsequent video segment, creators may use the final frame of the previous segment as an additional reference image.\n\nThe final frame should be provided as a primary continuity reference for the next sequence.\n\nThis approach improves:\n\n  * Character continuity\n  * Environmental continuity\n  * Camera continuity\n  * Motion continuity\n\n\n\nand reduces abrupt scene transitions between independently generated video segments.\n\n#### 4-1 Known Limitations\n\nWhen portions of a character move outside the camera view, those regions become visually undefined to the model.\n\nAs a result, when the hidden regions re-enter the frame, the AI may unintentionally alter:\n\n  * Clothing details\n  * Accessories\n  * Hair structure\n  * Body proportions\n  * Minor character features\n\n\n\nThis occurs because the model must reconstruct information that was not visible in the inherited reference frame.\n\n#### 4-2 Correction Method\n\nWhen continuity errors appear, creators may:\n\n  1. Capture a frame immediately before the error occurs.\n  2. Reintroduce the original character reference images.\n  3. Regenerate the affected segment.\n  4. Continue subsequent video generation from the corrected output.\n\n\n\nThis iterative correction workflow allows creators to maintain long-form narrative consistency while minimizing cumulative visual drift.\n\n#### 4-3 Preventive Practices\n\nWhenever possible, important character features, costumes, equipment, and narrative objects should remain visible during scene transitions.\n\nReducing long-term occlusion can significantly improve continuity preservation and reduce regeneration requirements.\n\n* * *\n\n### 5. Method Objective\n\nThe Hinoko Space Method addresses three primary consistency challenges commonly found in AI-assisted visual generation:\n\n  * Spatial Consistency\n  * Character Consistency\n  * Temporal Consistency\n\n\n\nRather than focusing solely on prompt engineering, the method emphasizes structured reference guidance.\n\nThe objective is to provide creators with a practical framework for maintaining continuity across images, videos, and long-form AI-assisted storytelling projects.\n\n**End**\n\nGo to: codeberg\n\nSearch: Hinoko-Space-Method\n\nYou can view the detailed case study. Thank you for watching again.",
  "title": "Hinoko-Space-Method V0.0.2"
}