从画图的手到品味决策者 · AI 与视觉设计产业 · 2026-07From the drawing hand to the taste-maker · AI & visual design · Jul 2026

AI 没有替代设计,它把设计拆成了三层——中间那双画图的手被整层抽走,剩下的价值往两头跑:一头是「定义要什么」,一头是「判断对不对」 AI didn't replace design — it split it into three layers. The drawing hand in the middle was pulled out whole, and the remaining value ran to the two ends: «defining what» and «judging whether it's right»

最小单元＝「一次看起来对的判断」(a taste call)。整条视觉生产链,本质是无数次「生成→看→判断对不对→再生成」的循环。过去这个判断动作埋在执行劳动里看不见;AI 把「画出来」的成本压到近零,判断被从执行中剥离出来、单独暴露在聚光灯下——「会选的人」成了新的稀缺品。The atomic unit is «a taste call». The whole visual pipeline is endless loops of «generate → look → judge → regenerate». That judgment used to hide inside execution labor; AI drove the cost of «making it» near zero, and judgment was stripped out of execution and lit up on its own — «the one who can choose» became the new scarcity.

执行成本归零 ≠ 决策成本下降。生成 40 个方案和生成 1 个一样便宜之后,瓶颈立刻换位:从「能不能做出来」变成「能不能选出对的那个」——「选择爆炸」让决策更难,不是更容易。也因此乙方的底牌露了出来:它卖的从来不只是图,是「确定感与背锅权」。Execution cost → 0 ≠ decision cost down. Once 40 options cost the same as 1, the bottleneck swaps instantly: from «can we make it» to «can we pick the right one» — the «explosion of choice» makes deciding harder, not easier. And the vendor's hole card shows: it never just sold images, it sold «certainty and someone to blame».

骨架是一个哑铃(barbell):横轴铺开视觉资产生命周期八节点(brief→调研→概念→执行→改稿→交付→资产→品牌),中段被掏空、两端鼓起。另含零号病人(2023 原画师裁员潮)、执行贬值 vs 判断升值的 K 型分化、六块硬骨头、龙头四路线、产品指南(2C 个人创作 / 2B 团队企业)。这是一张批判性行业解剖,不是工具测评。相邻议题见姊妹图:审美 IP 变现 / 设计师转自媒体→creator、设计转代码「生成 vs 验证」→code、设计教育转判断力→edu。 The skeleton is a barbell: the visual-asset lifecycle runs across eight nodes (brief→research→concept→execution→revisions→delivery→assets→brand), the middle hollowed out and the two ends bulging. Plus Patient Zero (the 2023 game-illustrator layoffs), the K-shaped split of execution-devalued vs judgment-appreciated, six hard bones, four leader routes, and a product guide (2C solo creators / 2B teams & enterprises). A critical dissection, not a tool review. Adjacent topics on siblings: taste-IP monetization / designers turning influencer → creator, design-to-code «generate vs verify» → code, design education toward judgment → edu.

执行层 · 被抽掉的手Execution · the pulled hand

brief 端 · 定义要什么Brief · defining what

taste 端 · 判断对不对Taste · judging if right

AI 生成 · 工具AI generation · tools

30–66%

2023 前三个月游戏外包公司原画师裁员比例(零号病人;广州某公司效率 3 倍、15 人→5 人、月省 20 万)。⚠️个案区间·非全行业Game-outsourcing concept-artist layoffs in Q1 2023 (Patient Zero; one studio hit 3× efficiency, 15→5 staff, ¥200k/mo saved). ⚠️Case range, not industry-wide

15%→<9%

自由职业平台入门级项目占比——「新手村」被摧毁、行业的第一级台阶被拆走;基础平面订单 −17%、基础文案 −30%Entry-level projects on freelance platforms — the «newbie village» razed, the industry's first rung removed; basic graphic orders −17%, basic copywriting −30%

+44%

AI 相关项目自由职业者时薪溢价(Upwork;平台 AI 费率一年 +60%)——同一个市场,一端在涨价、一端在归零:极端 K 型分化The hourly premium for AI-related freelance work (Upwork; platform AI rates +60% in a year) — one market, one end repricing up while the other zeroes out: an extreme K-shaped split

32%

只有 32% 创作者认为可依赖 AI 输出本身(Figma 2025,尽管 78% 承认提效)——效率与可依赖性之间的剪刀差。Adobe:81% 认为品味仍需人类判断Only 32% of creators feel they can rely on AI output itself (Figma 2025, though 78% grant efficiency) — the speed-vs-dependability scissors. Adobe: 81% say taste still needs human judgment

口径警告:本页是批判性行业分析,非工具测评 / 投资建议。裁员比例「30%–66%」是 2023 个案区间(引中青在线 / 澎湃等一线报道),非全行业统计;「插画岗一年降约 70%」(杭州招聘者口径)是多因叠加(2021 游戏版号冻结 + 经济放缓 + AI)、非官方;报价「跌到 1/10」为受访者个案陈述。自由职业平台数据(−17%/−30%/+44%)为第三方口径(C),学术锚《Who is AI Replacing?》(Management Science,61 国近 200 万招聘)图像岗 −17%、写作 −30%(A)。Adobe 2026 / Figma 2025 为厂商自办调查(含立场)但共同指向「效率↑、可依赖性↓」剪刀差。各公司估值 / 营收:Figma IPO 定价(A)、Midjourney 营收(B/D 非财报)、Canva 披露(D)逐个角标。Rayark「六指事件」标争议·未证实。Midjourney 的用户规模是 D 级:流传最广的「2025 年 3 月约 2100 万 Discord 注册、日活 120–250 万、25–34 岁约 37%、年收入约 5 亿美元」四个来源无一手——两家站点自述为「基于 Discord 服务器数据、订阅价格与用户行为的综合估算,算法未公开」,行业评论未给原始数据链接;可用于说明量级,不能用来算份额或做对比。注意方法论陷阱:Discord 注册数不是产品用户数,更不是付费用户数。每张卡片右上角 A/B/C/D＝证据强度;产品指南非推荐背书。 Basis warning: a critical industry analysis, not tool-review / investment advice. The «30%–66%» layoff range is a 2023 case span (from front-line reports), not an industry statistic; «illustration roles down ~70% in a year» (a recruiter's figure) is multi-causal (2021 game-license freeze + downturn + AI), unofficial; the «quotes cut to 1/10» is one interviewee's account. Platform figures (−17%/−30%/+44%) are third-party (C); the academic anchor «Who is AI Replacing?» (Management Science, ~2M postings across 61 countries) puts image roles at −17%, writing −30% (A). Adobe 2026 / Figma 2025 are vendor-run surveys (with a stance) but both point to the «speed↑, dependability↓» scissors. Company valuations/revenue: Figma's IPO pricing (A), Midjourney's revenue (B/D, non-filing), Canva's disclosures (D) are badged individually. Rayark's «six-finger incident» is marked disputed·unverified. Midjourney's user scale is grade D: the widely repeated «~21M Discord registrations by March 2025, 1.2–2.5M daily actives, ~37% aged 25–34, ~$500M annual revenue» rests on four non-primary sources — two sites describe their numbers as composite estimates from Discord server data, subscription prices and user behaviour with the method undisclosed; the industry commentary links no primary data. Usable for order of magnitude, not for share calculations or comparisons. Mind the methodological trap: Discord registrations are not product users, still less paying users. Each card's top-right A/B/C/D = evidence strength; the product guide is not an endorsement.

◆ 核心骨架 · 价值向两端坍缩The core skeleton · value collapses to two ends

一个哑铃:两端鼓起,中间被掏空A barbell: two ends swell, the middle hollows out

全图的脊椎就在这一节。传统视觉链里占绝对大头的「执行层」——那双画图的手——正被 AI 一整层抽走;价值没有消失,它搬家了:一头搬去定义要什么的 brief,一头搬去判断对不对的 taste。通用执行技能在贬值,「定义问题」与「判断什么图是对的」在升值——这不是预测,是已经发生的重新计价。The map's spine sits in this section. The «execution layer» that dominated the traditional pipeline — the drawing hand — is being pulled out whole by AI; the value didn't vanish, it moved house: to the brief that defines what, and the taste that judges whether it's right. Generic execution devalues; «defining the problem» and «judging which image is right» appreciate — not a forecast, a repricing that has already happened.

brief 端Brief end

定义要什么 ↑define what ↑

商业理解 · 抽象问题转译 · 定方向 / 风格 / 边界。「先写清楚要什么」比「会不会做图」更贵。Business sense · translating the abstract · setting direction/style/scope. «Writing what you want clearly» beats «being able to make it».

执行层Execution

↓

绘制 · 修图 · 矢量化 · 排版 · 渲染 · 剪辑——被系统性抽掉的手。Drawing · retouching · vectorizing · layout · rendering · editing — the hand, systematically pulled out.

taste 端Taste end

判断对不对 ↑judge if right ↑

审美直觉 · 文化洞察 · 时代情绪 · 验收 / 一致性维护。「会选的人」成为新稀缺枢纽。Aesthetic instinct · cultural insight · zeitgeist · acceptance / consistency. «The one who can choose» becomes the new scarce hub.

① 执行归零 ≠ 决策下降① Zero exec ≠ cheaper decisions

效率与可依赖性之间有一道剪刀差(下方数据)——做图更便宜了,拍板反而更贵了。A scissors gap between speed and dependability (data below) — making got cheaper, deciding got dearer.

② 改稿归零,暴露需求方② Free revisions expose the client

「过去一天画一个场景,现在一天出 40 个供老板挑」——瓶颈从「能不能做」换成「能不能选对」。«A day for one scene before; now 40 a day for the boss to pick» — the bottleneck swaps from «can we make it» to «can we pick right».

③ 选择爆炸,决策更难③ Choice explosion, harder decisions

面对成百上千张完美无瑕的图,客户陷入决策瘫痪——「会选的人」成为新稀缺。Facing hundreds of flawless images, clients hit decision paralysis — «the one who can choose» becomes scarce.

▸ 效率:AI 让生产更快Speed: AI makes production faster

93%Adobe 2026:AI 让生产更快Adobe 2026: AI makes production faster

78%Figma 2025:承认 AI 显著提效Figma 2025: AI meaningfully boosts efficiency

75%Adobe:已把 AI 视为「已整合或不可或缺」Adobe: AI is «integrated or essential»

◈ 可依赖性:但「对」还得靠人Dependability: but «right» still needs humans

32%Figma:仅 32% 认为可依赖 AI 输出本身Figma: only 32% can rely on AI output itself

57%Adobe:输出仍需中度 / 重度编辑才能发布Adobe: output needs moderate/heavy edits to ship

81%Adobe:创意品味仍需人类判断Adobe: taste still needs human judgment

「好看的通胀」:当所有人都能 3 秒生成「大师级」图,稀缺的就从「好看」换成了「对」——对品牌、对语境、对时代情绪。(Adobe:53% 因内容爆炸更难脱颖而出)«The inflation of beauty»: when anyone makes a «masterpiece» in 3 seconds, the scarce swaps from «pretty» to «right» — for the brand, the context, the zeitgeist. (Adobe: 53% find it harder to stand out.)

◆ 判断层杠杆排序 · 三份研究高度一致(结论卡)The leverage order · all three studies agree

1定义问题的能力＞ ~~执行能力~~Ability to define the problem ＞ ~~execution skill~~

2个人可识别的风格资产＞ ~~通用技能~~A recognizable personal style ＞ ~~generic skill~~

3审美积累与参考库＞ ~~工具熟练度~~Aesthetic depth & a reference library ＞ ~~tool proficiency~~

⚠️注:个人可识别风格是「危险而迷人的硬币」——它是最后的护城河,却也一旦确立就极易被提取、训练成 LoRA 模型无限克隆。护城河与泄洪口,是同一条河。⚠️Note: a recognizable personal style is a «dangerous, alluring coin» — the last moat, yet once established it is easily extracted and cloned into a LoRA model. The moat and the spillway are the same river.

从这张图看到的五条规律Five patterns this map makes visible