受众:产品所有者(vibe coder)。目的:把「古诗 × 绘画 × 生图模型 × MDA 游戏化」四件事揉成可落地的设计决策。 日期:2026-05-25 · 模式:deep(8 阶段)· 来源:112 条,覆盖游戏设计/美术史/HCI/产品复盘四类。
产品已用 MDA 框架锁定核心美学为发现(Discovery)——让普通人惊讶于「同一首诗竟能被画成这么多种样子、绘画原来有这么多门道」,而非自我表达(Expression)。本研究围绕这个已定前提,回答「怎么设计才能兑现发现」。
四条最尖锐的结论:
你的产品形态本身就是一套被实证有效的视觉素养教学法。 美术教育界的 VTS(视觉思维策略)证明:让人学会「看画」最有效的方式不是讲术语,而是把同一对象的不同处理并排放出来,让对比把「作者的选择」逼现出来 [41][42][43][44]。中国画史早有同构传统——王维《辋川图》被历代反复重画、北斋《富岳三十六景》同一座山三十六种构图、塞尚同一座圣维克多山画了约 80 遍 [26][47][45]。「一诗多画」不是噱头,是千年范式 + 教育学正解。
「哇」的强度由维度决定,且可排序。 最能让外行一眼震惊的是画种/媒材(工笔 vs 写意 vs 青绿 vs 油画,肉眼一秒可辨「换了物种」),其次是景别(贴脸特写 vs 天地间一小人)和色彩冷暖(暖橙夕照 vs 冷蓝月夜);散点透视、留白属于「配一句话讲解就豁然开朗」的第二层惊喜 [28][29][32][36][30]。这告诉你:要制造发现,先变这三样。
150–180 秒的慢渲染 + 不可重来,是资产不是成本——前提是被重新包装。 多巴胺在「期待时」达峰而非「获得时」;对相信结果是好的正面事件,人愿意「品味式地等」;峰终定律说人只记住「最强的一刻和结尾」[68][69][80]。把等待做成「显影/开盲盒」仪式,把揭晓做成产品的「峰」,慢就变成体验主线。而 Wordle 式不可重来恰好是对抗抽卡氪金暗黑模式的护城河,同时让每次揭晓成为真赌注 [11][82][112]。
头号死法是变成另一个妙鸭相机。 妙鸭 2023 年爆火后一年下载量跌 98%、2025 年团队解散,死因是「低频工具 + 用完即走 + 审美疲劳 + 无社区」[106][108]。纯「输入诗→出一张好看图」必死。解药是研究反复指向的同一组东西:发现 + 收集 + 每日 + 做同款(社区)。
基于此,报告给出三个可分别做实验的设计方向(盲盒一笔定 / 一诗百象图鉴 / 视觉旋钮),以及六条贯穿所有方向的设计铁律。三个方向不互斥,可叠加;推荐先用「盲盒一笔定」做最快验证。
范围内:以发现为核心乐趣的交互设计、机制设计、等待与揭晓体验、绘画维度的选择、可借鉴的同类产品。
范围外(按已定前提,不在本研究内重新论证):是否该选发现而非表达(已定)、是否保留慢渲染与不可重来(物理约束 + 已定)、具体技术栈与存储实现。
方法:deep 模式四领域并行检索——(1) MDA 与发现型游戏机制;(2) 绘画视觉语言与视觉素养教育;(3) 文生图模型能力边界与慢生成/揭晓 UX;(4) 同类产品复盘。共采集 112 条来源,跨领域三角验证。
关键假设(高材料性,显式声明): - 假设产品仍使用类 gpt-image-2 的高质量慢模型(150–180s/张),未引入快预览模型。方向三对这条假设最敏感,下文专门标注。 - 假设目标用户是「对古典文学有兴趣的成年人」,非应试学生、非专业设计师(与 PRD 一致)。 - 假设产品保留「我要选/我要写」的有意义选择(与既有产品决策一致),不退化为用户纯旁观。
MDA 的铁律:设计师只能直接控制机制(Mechanics),乐趣(Aesthetics)是经由动态(Dynamics)间接涌现的——你写不出一行叫「发现感」的代码,只能造机制、跑出动态、让发现涌现 [1]。LeBlanc 的「八种乐趣」里,发现(Game as uncharted territory)与表达(Game as self-discovery)相邻但本质不同:表达是「通过创作认识自己」,发现是「探索外部未知疆域」[2]。这条区分直接决定产品叙事——不该说「画出你心里的诗」,该说「看看这首诗还能是什么样」。
底层只有两台心理引擎,其余机制都是它俩的变体:
由此得出的可迁移机制清单(每条都附心理学出处,建议逐条当实验项):
| 机制 | 怎么运作 | 出处 |
|---|---|---|
| 渐进揭示 | 先给启动剂信息开缺口,再分次填补;每次填补既是奖赏又开新缺口 | [3][2] |
| 盲盒/揭晓仪式 | 翻牌「前」的不确定比结果更刺激;揭晓动画的过程才是高潮 | [8][9] |
| 收集/完成主义 | 「为收集而收集」——进度条、完成度、图鉴把「看遍所有可能」变成可量化目标 | [14] |
| 可控变异 | 同一输入沿不同维度演绎,产出各异且都体面,意外即乐趣 | [17][18] |
| 并置对比 | 同屏并排多种结果,制造「原来还能这样」的认知冲突——发现体感的核心载体 | [2][4] |
| 杂交/混合 | 把两种解读杂交出第三种没想到的中间态 | [17] |
| 首次发现署名 | 「你是第一个把这首诗画成这样的人」,把单机生成变成全球共享实验 | [95][96] |
随机与控制怎么调(最关键的调参结论):好的随机放在输入侧(在用户做选择「之前」改变局面,用户仍能基于新局面做有意义决策),坏的随机放在惩罚侧(你做对了却因 RNG 出烂图)[19]。保底/pity 机制给随机加「地板」,用完成主义对冲挫败 [21][10]。对你:同一首诗交给不同风格演绎使结果各异(输入侧随机,good),但要用策展过的风格库做质量地板,绝不让不可重来的一笔撞上一张明显跑偏/劣质的图——否则发现瞬间变惩罚。
把一首诗「画出来」,可换的视觉变量很多,但对非专业用户的冲击力天差地别。综合美术史、影视摄影语言、中国画论与视觉素养研究,按「一眼可感 + 落差大」排序:
产品启发:第一组对比就主打 画种 ×(景别 / 冷暖),落差最大、最不用解释;散点透视、留白留作「配一句话就豁然开朗」的第二层。
文化与教育学背书(这点决定产品合法性与方法): - 「诗中有画,画中有诗」出自苏轼评王维,代表中国画从「模拟形似」转向「表达心意」——写意精神的源头 [22][23][25]。写景、有鲜明视觉意象的诗最可画;纯抒情说理的难画 [24]。 - VTS(视觉思维策略):Housen 录下几千次观众看画反应,发现理解艺术按可预测的「阶段」推进,而绝大多数观众是「初学者观者」;传统讲解式导览对多数人无效。VTS 的做法是引导者不直接给答案,用固定几个问题引导观众探索,长期内化出独立「看」的能力,并被证明能提升审美思维与依证推理 [41][42][43]。你的产品天然就是一台 VTS 机器——不灌术语,靠并置同一首诗的 N 种画法,让「作者的选择」自己浮出来。这与你既定的产品内核「审美靠判断角度、不靠术语」完全同向。
模型能力的硬事实(落差的技术根源):基准 GenEval 2(2025)显示,原子级(单物体/单属性)最强模型能到人类判分 85.3%,但整句提示级(多约束同时满足)最强模型只有 35.8%[55]。也就是提示越长、约束越多,模型同时全部命中的概率断崖下跌。多物体多属性时会张冠李戴、漏物体、搞错空间关系 [56]。
两类模型的分工:GPT-Image 类遵循度高、能把整段汉字写进画(约 98% 文字准确);Midjourney 类「演绎多于执行」,像艺术总监用氛围光、电影感填补提示空白——这种「创造性漂移」正是惊喜的最佳引擎,但汉字必翻车 [52][54]。含义:要把诗句写进画面就用 GPT-Image 类,或把诗句作为界面排版叠在画外、不让模型画字。
「AI 味」平庸是惊喜的头号杀手:RLHF 对齐导致 mode collapse(模型收敛到少数安全典型答案),这是很多 AI 图「一眼 AI、毫无惊喜」的根因 [64]。对策 Verbalized Sampling 思路:内部多生成候选、挑「非典型但贴题」的那张揭晓,多样性可提升 1.6–2.1×[64][65]。
好惊喜 vs 坏落差的判据:好惊喜=输出明显「是我要的那首诗」,但多给了我没想到的氛围/光/细节;坏落差=丢了我明确点名的核心元素、出现诡异(uncanny)、或「输入变了输出没变」(系统没在听我说话)[60][61]。
等待与揭晓的体验模式(直接拿去用): - 显影/宝丽来隐喻:把 150–180 秒做成「照片在显影」,模糊→清晰逐步浮现,而非转圈。Flashback app 故意让你等 24 小时看照片,证明「故意慢」能做成卖点 [71][72]。 - 把扩散去噪过程做成可看的演进:Midjourney 的 blurry→sharp 本就是去噪可视化,天然契合「显影」[59]。 - 骨架屏式预告:先把画框、诗句排版的位置占好,制造「这里即将出现画」的期待,实测比纯进度条更快更愉快 [73][74]。 - 分层揭晓:一次只露一层(先色调、再轮廓、再细节,或先画后诗),把单次揭晓拆成多次小奖赏 [79]。 - 把揭晓当峰终时刻重点投入:配庆祝动画/音效/微交互,资源优先级最高 [80][81]。 - 给等待赋予意义:等待时呈现这首诗的背景、为什么这样画的小线索,把空等变成「品味」[69][75]。
你最像的三个产品的杂交体:Artle + Infinite Craft + DailyArt。
必须躲开的坑: 1. 别做成另一个妙鸭/Lensa:纯工具=低频、审美疲劳、用完即走 [106][108][110]。靠发现+收集+每日+做同款续命。 2. 别只给图不给「看的角度」:每日故宫被批「大量文物没介绍、核心体验没做细」[97]。判断角度要短、不像上课,但必须有。 3. 别把「再画一次」滑向抽卡赌博:可变奖励+付费重抽=暗黑模式 [112]。Wordle 式低迭代约束反而是你的天然护城河,要守住。
无论选下面哪个方向做实验,这六条都成立——它们是研究收敛出的「不要踩」:
三者不互斥,可叠加。每个都按 MDA 三层 + 与慢渲染的契合度 + 与现有代码的距离来描述。
一与二可叠加:图鉴的每一格,用「盲盒揭晓」的方式来填。这就是 Artle(每日揭晓)+ Pokédex(收集图鉴)的合体。
游戏设计 / MDA / 发现心理学 1. MDA framework — Wikipedia — https://en.wikipedia.org/wiki/MDA_framework 2. The 8 Kinds of Fun — Skeleton Code Machine — https://www.skeletoncodemachine.com/p/the-8-kinds-of-fun 3. Curiosity, Information Gaps, and the Utility of Knowledge — Golman & Loewenstein, CMU — https://www.cmu.edu/dietrich/sds/docs/golman/golman_loewenstein_curiosity.pdf 4. The psychology and neuroscience of curiosity — PMC — https://pmc.ncbi.nlm.nih.gov/articles/PMC4635443/ 5. (Don’t) mind the gap? Information gaps compound curiosity yet feed frustration — ScienceDirect — https://www.sciencedirect.com/science/article/abs/pii/S0749597823000523 6. Variable Rewards: Want To Hook Users? — Nir Eyal — https://www.nirandfar.com/want-to-hook-your-users-drive-them-crazy/ 7. 3 Types of Variable Rewards — Nir Eyal (LinkedIn) — https://www.linkedin.com/pulse/3-types-variable-rewards-hook-your-users-nir-eyal 8. The Compulsive Nature of Gacha Games — Medium — https://medium.com/@kedas_38457/the-compulsive-nature-of-gacha-games-55016c7b54da 9. Loot Boxes, Gacha, And The “Near-Miss” Effect — GeekVibesNation — https://geekvibesnation.com/loot-boxes-gacha/ 10. The Science of Gacha: Drop Rates and Fairness — https://loot-box-probability-mechanics.pages.dev/blog/the-science-of-gacha-understanding-drop-rates-and-fairness 11. The Psychology of Wordle — Psych News Daily — https://psychnewsdaily.com/the-psychology-of-wordle/ 12. Roguelike Level Design Addendum — Grid Sage Games — https://www.gridsagegames.com/blog/2019/03/roguelike-level-design-addendum-static-procedural/ 13. Why Are Roguelike Games So Engaging — RetroStyle Games — https://retrostylegames.com/blog/why-are-roguelike-games-so-engaging/ 14. Creating the Craving: Why is There a Pokedex? — Treasure Savvy — https://treasuresavvy.wordpress.com/2016/04/10/creating-the-craving-or-why-is-there-a-pokedex/ 15. Townscaper — Wikipedia — https://en.wikipedia.org/wiki/Townscaper 16. How Townscaper Works — Game Developer — https://www.gamedeveloper.com/game-platforms/how-townscaper-works-a-story-four-games-in-the-making 17. Artbreeder — Wikipedia — https://en.wikipedia.org/wiki/Artbreeder 18. Exploring Latent Dimensions of Crowd-sourced Creativity — arXiv — https://arxiv.org/pdf/2112.06978 19. Randomness and Game Design — Game Developer — https://www.gamedeveloper.com/design/randomness-and-game-design 20. When Randomize Game Design Goes Too Far — Game Wisdom — https://game-wisdom.com/critical/randomness-game-design 21. How to Fight RNG in Game Design — Game Developer — https://www.gamedeveloper.com/design/how-to-fight-rng-in-game-design
绘画语言 / 诗画传统 / 视觉素养 22. 诗中有画画中有诗 — 百度百科 — https://baike.baidu.com/item/诗中有画画中有诗/201242 23. 诗中有画,画中有诗 — 百度百科 — https://baike.baidu.com/item/诗中有画,画中有诗/8621409 24. 王维经典诗八首 — 澎湃新闻 — https://www.thepaper.cn/newsDetail_forward_12709488 25. 诗歌与绘画——从王维看中国古代的诗画论 — 新浪博客 — https://blog.sina.com.cn/s/blog_1989145ad0102ynki.html 26. 宁晓萌:辋川图现象 — 北京大学人文社会科学研究院 — http://www.ihss.pku.edu.cn/templates/zs_mt/index.aspx?nodeid=152&page=ContentPage&contentid=2544 27. 王维与辋川别业 — 上海老年报 — https://mzj.sh.gov.cn/lnb-wsws/20220208/b797a141b97e486dacb33313c4ec6144.html 28. 国画中的34种技法 — 网易 — https://www.163.com/dy/article/GCN3ND8A0521BOGU.html 29. 长安雅士浅论工笔画与写意画的区别 — 知乎 — https://zhuanlan.zhihu.com/p/441276229 30. 浅析中国山水画中的散点透视和意境美 — 中国收藏网 — http://news.socang.com/2019/08/20/1712352402.html 31. 中国画散点透视和油画焦点透视有何不同 — 百度百科TA说 — https://wapbaike.baidu.com/tashuo/browse/content?id=2792313e21aaa0fb6810e8cc 32. Guide to Camera Shots: Every Shot Size Explained — StudioBinder — https://www.studiobinder.com/blog/types-of-camera-shots-sizes-in-film/ 33. 16 Types of Camera Shots & Angles — Boords — https://boords.com/blog/16-types-of-camera-shots-and-angles-with-gifs 34. Chiaroscuro — Wikipedia — https://en.wikipedia.org/wiki/Chiaroscuro 35. How to Use Chiaroscuro — MasterClass — https://www.masterclass.com/articles/how-to-use-chiaroscuro-to-add-dimension-to-your-film 36. Creating Mood in Art with Color Temperature — Nuart Planet — https://nuartplanet.com/creating-mood-in-art-with-color-temperature-warm-cool-explained/ 37. The Power of Color: Art’s Emotional Palette — WikiArt — https://www.wikiart.org/news/exploring-the-role-of-color-in-art-and-emotion/ 38. What Is Negative Space? — Skillshare — https://www.skillshare.com/en/blog/what-is-negative-space-design-psychology/ 39. Negative Space in Art — Zephyeer — https://zephyeer.com/blogs/art-education/negative-space-in-art 40. How to Build Visual Literacy — Serenade Magazine — https://serenademagazine.art/how-to-build-visual-literacy-a-beginners-guide-to-looking-at-art/ 41. What is VTS? — The Thinking Eye — https://www.thinkingeye.org/what-is-vts 42. Overview of Aesthetic Development — VTS Home — https://vtshome.org/aesthetic-development/ 43. Theory into Practice: Visual Thinking Strategies — Philip Yenawine — https://www.philipyenawine.com/vts/2020/8/23/theory-into-practice-visual-thinking-strategies 44. Art Pedagogy — Jerwood Visual Arts Glossary — https://jerwoodvisualarts.org/art-education-and-methodologies-glossary/art-pedagogy/ 45. Cézanne Painted Mont Sainte-Victoire Dozens of Times — Artnet — https://news.artnet.com/art-world/cezanne-mont-sainte-victoire-1937995 46. A Closer Look at the Mont Sainte-Victoire Series — Draw Paint Academy — https://drawpaintacademy.com/mont-sainte-victoire/ 47. Thirty-six Views of Mount Fuji — Wikipedia — https://en.wikipedia.org/wiki/Thirty-six_Views_of_Mount_Fuji 48. Hokusai’s 36 Views & Japonisme — GalleryThane — https://gallerythane.com/en-us/blogs/news/hokusais-36-views-of-mount-fuji-its-influence-on-western-art-and-japonisme 49. 36 Views of the Golden Gate Bridge — Arthur Drooker — https://www.arthurdrooker.com/the-bridge/
文生图模型 / 期待落差 / 等待与揭晓 UX 50. What is the Best AI Image Generator — CometAPI — https://www.cometapi.com/what-is-the-best-ai-image-generator/ 51. What Is FLUX 1.1 Pro — MindStudio — https://www.mindstudio.ai/blog/what-is-flux-1-1-pro 52. GPT Image 2 vs Midjourney v7 (2026) — AI.cc — https://www.ai.cc/blogs/gpt-image-2-vs-midjourney-v7-comparison-2026/ 53. GPT Image 2 vs Midjourney — Skywork — https://skywork.ai/skypage/en/gpt-image-2-vs-midjourney/2047148805738819585 54. Midjourney Complete Guide 2026 — AI Video Bootcamp — https://aivideobootcamp.com/blog/midjourney-complete-guide-2026/ 55. GenEval 2 — arXiv 2512.16853 — https://arxiv.org/html/2512.16853v1 56. Investigating Prompt Engineering in Diffusion Models — arXiv 2211.15462 — https://arxiv.org/abs/2211.15462 57. Harnessing Seeds for Variations in Stable Diffusion — https://aiimagegenerator.is/blog-Harnessing-the-Power-of-Seeds-for-Variations-in-Stable-Diffusion-Images-1357 58. What is CFG Scale (Stable Diffusion) — AI Photo Generator — https://www.aiphotogenerator.net/blog/2026/02/what-is-cfg-scale-stable-diffusion 59. Quality — Midjourney Docs — https://docs.midjourney.com/hc/en-us/articles/32176522101773-Quality 60. UX of AI Art Generators: Magical, Mystifying, Macabre — UserTesting — https://www.usertesting.com/blog/ux-ai-art-generators-magical-mystifying-and-macabre 61. Default Images in Text-to-Image — arXiv 2505.09166 — https://arxiv.org/html/2505.09166v5 62. Agency in Human-AI Collaboration — Taylor & Francis 2025 — https://www.tandfonline.com/doi/full/10.1080/10400419.2025.2587803 63. Prompt Engineering for Stable Diffusion — Portkey — https://portkey.ai/blog/prompt-engineering-for-stable-diffusion/ 64. Verbalized Sampling — arXiv 2510.01171 — https://arxiv.org/abs/2510.01171 65. Verbalized Sampling explainer — DataSci Ocean — https://datasciocean.com/en/paper-intro/verbalized-sampling/ 66. User-centered Co-creativity Framework — ACM CHI — https://dl.acm.org/doi/fullHtml/10.1145/3613905.3650929 67. Agency-First Framework for Generative AI — MDPI Electronics — https://www.mdpi.com/2079-9292/15/4/877 68. Anticipation: The Psychology of Waiting in Line — psychologicalscience.org — https://www.psychologicalscience.org/news/were-only-human/anticipation-the-psychology-of-waiting-in-line.html 69. Impatience, Savoring vs Dread (Hardisty 2020) — J. Consumer Psychology — https://myscp.onlinelibrary.wiley.com/doi/abs/10.1002/jcpy.1169 70. The Science of Anticipation — Medium — https://medium.com/@kp_80900/the-science-of-anticipation-why-looking-forward-to-something-might-be-better-than-the-thing-itself-31fb53dc92ee 71. Polaroids, Delayed Gratification & the Dopamine Problem — Very Big Brain — https://verybigbrain.com/psychology-thinking/polaroids-delayed-gratification-and-the-modern-brains-dopamine-problem/ 72. Flashback (instant-film app) — https://joinflashback.co/en-us 73. How to Speed Up Your UX with Skeleton Screens — SitePoint — https://www.sitepoint.com/how-to-speed-up-your-ux-with-skeleton-screens/ 74. Skeleton loading screen design — LogRocket — https://blog.logrocket.com/ux-design/past-present-skeleton-screen/ 75. Designing Better Loading & Progress UX — Smart Interface Design Patterns — https://smart-interface-design-patterns.com/articles/designing-better-loading-progress-ux/ 76. The Psychology of Waiting in UX — Design Bootcamp — https://medium.com/design-bootcamp/the-psychology-of-waiting-in-ux-0f0b24cdeb8f 77. Unboxing the User Experience — Medium (Aditii) — https://medium.com/@aditis.singh99/unboxing-the-user-experience-why-that-pretty-box-matters-more-than-you-think-fde4c92bb82e 78. Designing the Unboxing Experience — The Good — https://thegood.com/insights/unboxing-experience/ 79. 5 Rules for the Unboxing Experience — crowdspring — https://www.crowdspring.com/blog/packaging-unboxing-experience/ 80. The Peak-End Rule — Nielsen Norman Group — https://www.nngroup.com/articles/peak-end-rule/ 81. Peak-End Rule — Laws of UX — https://lawsofux.com/peak-end-rule/ 82. Operant Conditioning in UI/UX — Design Bootcamp — https://medium.com/design-bootcamp/designing-user-delight-leveraging-operant-conditioning-in-ui-ux-for-engaging-experiences-4c3bb7d3b374
同类产品 / 复盘 83. Poem Postcards — Google Arts & Culture — https://artsandculture.google.com/experiment/poem-postcards/ZgG5_uTbBgTpIQ 84. Google’s Redesigned Arts & Culture App — TechCrunch — https://techcrunch.com/2023/08/09/googles-redesigned-arts-culture-app-ai-generated-postcards-feature-play-tab/ 85. 国产AI作画神器文心一格 — 量子位 — https://www.qbitai.com/2022/08/37097.html 86. AI绘画6大工具对比 — 优设网 — https://www.uisdc.com/6-ai-draw-tools 87. 即梦AI 官网 — https://jimeng.jianying.com/ 88. 即梦 Dreamina 介绍 — AI工具网 — https://www.ai138.com/link/1489.html 89. 数字故宫 — 故宫博物院 — https://www.dpm.org.cn/bottom/friend.html 90. 数字故宫/畅游多宝阁 — 数英 — https://www.digitaling.com/projects/152887.html 91. Artle — National Gallery of Art — https://www.nga.gov/artle 92. New Daily Game Like Wordle for Art — Smithsonian Magazine — https://www.smithsonianmag.com/smart-news/new-daily-game-like-wordle-for-art-180980083/ 93. Artle — The Art Newspaper — https://www.theartnewspaper.com/2022/05/26/artle 94. NYT Connections — Tom’s Guide — https://www.tomsguide.com/news/todays-connections-answer 95. Infinite Craft — Wikipedia — https://en.wikipedia.org/wiki/Infinite_Craft 96. Infinite Craft addictive loop — Oreate AI — https://www.oreateai.com/blog/infinite-craft-the-endlessly-addictive-game-that-lets-you-build-worlds-and-sometimes-wild-stories/d0f6a49afa549036abc158a8914f05c9 97. 每日故宫设计点评 — 少数派 — https://sspai.com/post/28953 98. 如何评价每日故宫 — 知乎 — https://www.zhihu.com/question/28174414 99. DailyArt — My Modern Met — https://mymodernmet.com/dailyart-app-zuzanna-stanska/ 100. DailyArt 官网 — https://www.getdailyart.com/ 101. Play with Google Arts & Culture — https://artsandculture.google.com/play 102. 4 New Games from Google Arts & Culture — Google Blog — https://blog.google/outreach-initiatives/arts-culture/4-new-games-and-experiments-from-google-arts-culture-to-inspire-your-summer/ 103. Artbreeder features — aitoolscoop — https://aitoolscoop.com/tool/artbreeder/ 104. Generative Art in NFT Minting — Block3 Finance — https://www.block3finance.com/the-role-of-generative-art-in-the-future-of-nft-minting 105. Cautionary Tales From Hyped NFTs — nftnow — https://nftnow.com/features/cautionary-tales-from-hyped-nfts-pixelmon-hape-prime-and-more/ 106. 妙鸭相机一年后复盘 — 知乎 — https://zhuanlan.zhihu.com/p/710339081 107. 妙鸭相机现状 — 澎湃新闻 — https://www.thepaper.cn/newsDetail_forward_28151694 108. 妙鸭团队解散 — 网易 — https://www.163.com/dy/article/KTFFCSPP0511U82T.html 109. Why Is AI Art So Cringe — Vice — https://www.vice.com/en/article/why-is-ai-art-so-bad/ 110. After the Hype: Why AI Art Will Fade — Craig Boehman — https://craigboehman.com/blog/after-the-hype-why-ai-art-will-fade-to-a-commercial-niche 111. Loot box — Wikipedia — https://en.wikipedia.org/wiki/Loot_box 112. Gambling Mechanics in Gacha Games — Uniwriter — https://www.uniwriter.ai/psychology/gambling-mechanics-in-gacha-games-their-detrimental-effects-and-potential-mitigation-measures/
search CLI 仅配 stealth
provider,子代理改用 WebSearch + WebFetch 完成。MDA 原始 PDF
抓取为二进制乱码,发现的定义以 Wikipedia + Skeleton Code Machine
二手交叉确认(两者一致:「Game as uncharted territory」)。