# A Chinese AI-design Bilibili creator turned one of his videos into a skill: 37 days, 2800 stars, and the stars came from his own audience · open-source GitHub project

> Source: https://stringers.io/en/github/gh_oil-oil__oil-ui.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> The author is a Chinese AI-community creator with a following. Videos and posts went up the day it launched, and developers such as Yu Gu (程序员鱼皮) and GitHubDaily reposted it on their own. Since then he has shipped nearly every day, with images and posts for each release. In late October it got another round of reposts from Korean, Persian and English design accounts.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

oil-oil/oil-ui · oil-oil

# A Chinese AI-design Bilibili creator turned one of his videos into a skill: 37 days, 2800 stars, and the stars came from his own audience

**What it is** A UI design skill for coding agents like Claude Code and Codex: first it produces several genuinely different design directions side by side, then refines them against real desktop and mobile screenshots.

**Why it took off** The author is a Chinese AI-community creator with an audience. He posted a video and a tweet on launch day, and FishSkin (程序员鱼皮), GitHubDaily and others reshared it on their own; afterwards he shipped almost daily with images in each post. In late October it got another round of reposts from Korean, Persian and English design accounts.

**C** **How you could follow it** Don't build yet another skill that makes AI-made UI look good; copy the combination: turn a method you've already taught into a step-by-step workflow, pair it with a gallery of works that include the prompts, apply it to the empty niches like slides, App Store screenshots and Chinese typesetting, and launch it with your own video.

Grade C: the stars mostly come from the author's own fans and his name in the Chinese-language community, which followers can't reproduce; UI design skills are the most crowded slot, with four or five giants at 100k+ stars, and unslop, which entered a month later, is growing faster. Being talked about in small circles

Data (updated daily) · Repo created 2026-09-30 · 2,908 stars · 7 days +355

GitHub · stars

Data 2026-10-09 → 2026-11-08 (31 days, 31 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

oil-ui is a UI design skill for coding agents like Claude Code and Codex. Its author oil has about 10k followers on X and 20k on Bilibili. On Sep 30 he posted a video and a tweet the same day; it passed 1,000 stars in the first week and went from 11-06 to 2,805 stars (GitHub). The method is somewhat new, but in its category the leaders have dozens of times its star count. Below: what it is, where the stars came from, who's around it, and what you can take from it.

### Sep 30 afternoon: an old video becomes a repo

2026 Sep 30, Beijing time 15:52: Bilibili creator 「oil欧呦」 posted a 7-minute 48-second video titled 《这个 Skill 帮你做出更好看的 UI 设计!》 (This Skill helps you make better-looking UI designs!). Two minutes later he posted the same news on X: "I previously posted a video on how to unlock AI's creativity to make non-mediocre designs; this time I organized the methods from it into a Skill." The post lists five steps: translate "premium" and "minimal" into concrete type, whitespace and color; start each direction from a concrete inspiration like a material or a scene; make the directions as different from each other as possible; compare them side by side on a built-in comparison page; after picking one, check it with screenshots on desktop and mobile, then remove things. He also added a line that cheap models like DeepSeek, mimo and gpt luna can produce decent results.

This post got 167 likes and 12.6k views, and the video reached about 5,700 plays (11-06). Not huge numbers, but it reached the right people: Chinese developers who use AI to write frontend code every day and are sick of the "AI-looking" page style. Early on the morning of Oct 6, he posted again: "In two days it has 570 stars... Turns out a lot of bloggers recommended this Skill on their own." By 10-07, his posts were already saying "nearly 1k stars."

Author Zhihuang Lin (GitHub oil-oil, 1,634 followers) makes AI tool videos and posts for a living, nearly every day. This isn't his first project: his draw-ui earned 544 stars, and oil-ui is the second-generation rebuild. Under the same account he has 30 skills and tool repos totaling about 13k stars, with oil-motion at 2.8k and beautify-github-readme at 1.95k.

### How it differs from other UI skills

"You can tell at a glance that AI made this page": purple gradients, a centered big headline, three rounded cards. Nearly every post that reshared it opened with this line. Most skills in the market that address this use the same approach: stuff the model with a taste ruleset or an "anti-AI-look" checklist and have it produce one option at a time.

oil-ui changes the process instead: first break down the category and find references, produce several very different directions at once and pick among them on a comparison page, then review desktop and mobile screenshots, and finally strip things back. This difference can be stated in one sentence, and reposting accounts summarize it that way most often. There's also a works gallery, where each piece comes with its prompt and the cheap model used, which reposting accounts pulled straight in as images.

But don't overrate this difference. The next section shows that it isn't what actually drove the stars.

### How the stars came

| Timeline | Stars | What happened |
|---|---|---|
| 09-30 | 0 | The author's X post and Bilibili video launched the same day |
| 10-04 ~ 10-06 | ~570 → nearly 1,000 | Reposted by 程序员鱼皮 10-04 times (17.8k views), and by GitHubDaily 10-06 times; the author's two update posts got about 66k views each, 10-03 and 10-06 |
| 10-09 | 1,085 | Over 1,000 in the first week |
| 10-24 | 2,058 | The author shipped almost daily, posting a new work image each time |
| 10-26 ~ 10-27 | 2,229 → 2,364 | Daily +138, +135, right after the Persian AI account @Yarad_ai's "passed 2 stars" post (16k views) on 10-25 |
| 11-05 | 2,748 | +76, the day after the Korean account @dopamine_crypto posted "2,208 stars in a month" (16k views) |
| 11-06 | 2,805 | Last 7 days +309, the 7 days before that +481. It's cooling. |

Star counts come from GitHub's daily numbers (starting 10-09); the figures from late September to early October are the ones the author gave in his own posts. It never made Hacker News, and there's no sign of paid promotion. Broken down, four forces drove it:

**The author's own audience.** The launch-day video and tweet, plus two update posts at 60k+ views in the first week, pushed the repo to the 30k people who already followed him. This is the biggest force, and the one followers can't take.

**Chinese recommendation newsletters relay it.** Yupi, GitHubDaily, GitTrend, plus reposts from 苏米客, AI拍档 and Zhihu columns all picked it up within the first month. They repost because the authors already have a name in this circle.

**Ship every day, post every time.** 37 days in, now at v0.32.0,10-11, and on one day alone it shipped 5 versions. Each release goes out with a fresh artwork image on X, and every few days that creates a small spike.

**The second wave of foreign-language reposts.** After the Chinese-language buzz faded in late October, the English design account Nick Babich (11-05 post listed it first, 11.7k views), plus Persian and Korean accounts, each brought a small second peak. The author also built a site in 6 languages with 870 pages, including a head-to-head comparison with Anthropic frontend-design, UI UX Pro Max and Impeccable, riding on the big names' search terms. How much traffic this brought is not visible from outside.

Then DeepSeek Harness dropped, and he built a native plugin version, dsh-oil-ui (214 stars), in 10-18.

Another number worth noting: oil-ui's cumulative installs on skills.sh are 3.3K (11-06), while Impeccable on the same page has 1.9M. The stars are decent, but few people are actually using it.

### Who's nearby

| Project | Launched | Scale (11-06) | How oil-ui differs |
|---|---|---|---|
| Anthropic frontend-design | 2025-09 | Parent repo 188k stars, 3.7M installs | The official general aesthetic guide: single direction, ready out of the box |
| nextlevelbuilder/ui-ux-pro-max-skill | 2025-11 | 146k stars, 910k installs | A large rule library for style, color and type |
| Leonxlnx/taste-skill | 2026-02 | 111K stars | A "taste" rule pack, single direction |
| pbakaus/impeccable | 2025-11 | 104k stars, 7 days +8,787; 1.9M installs | A set of design commands for polishing and reviewing existing UI |
| Nutlope/hallmark | 2026-04 | 49K stars | Anti-AI-slop rules, by a big-account author |
| alchaincyf/huashu-design | 2026-04 | 29K stars | A multi-style design skill from a Chinese author, the closest Chinese rival |
| heydamir/unslop | 2026-10-29 | 9,647 stars | Launched a month later, it turns vibecoded pages into 'designed' ones, and its star count grew more than 8x faster than the 3 days of oil-ui |
| Nutlope/variate | About 2026-07 | 221 stars | Generates 4 variants of the same component; a predecessor of side-by-side option comparison, but only at component level |

UI design skills are the most crowded cell among agent skills. variate already did side-by-side multi-direction comparison, yet has only 221 stars; unslop's one-line 'remove AI slop' passed oil-ui's star count in 8 days, when oil-ui took 37. In this cell, stars are decided by the author's reputation and how blunt the pitch is. How well the workflow is designed comes second.

### If it were you

**C: Learn from it, but don't go head-on.** oil-ui's 2,800 stars came mainly from its author's 30k followers and Chinese recommendation accounts, which followers can't replicate. In its cell, four or five large-scale repos with hundreds of thousands of stars already exist, and unslop, which entered a month later, is gaining faster.

**Who it suits** : Developers who already have an audience in some niche, especially those making tutorial videos or teaching methodology, who have a good piece of content they haven't turned into a tool yet, or who know a vertical area where no one has built a good skill yet.

**Who it doesn't suit** : Anyone with no distribution channel planning yet another 'make AI produce good-looking UI' skill. In this cell, rule libraries, command sets and multi-option comparison have all been done, and big accounts are already doing them.

**How to build it** : Copy its combination, not its topic. Turn a methodology you've already taught or written up into a skill, but don't ship a plain rules list. Build a workflow with steps that shows intermediate results, and add a showcase gallery where every piece is labeled with its prompt and model. Move the topic beyond UI into cells that are still empty: slide decks and data charts, App Store screenshots, email templates, admin pages for a specific framework, or Chinese typesetting, which the big players don't cover. Before launch, search skills.sh and the various skill directories. If a same-topic skill already has tens of thousands of stars, pick another topic.

**How to get users** : On day one, launch with your own videos and posts. Videos matter more than the README. After each release, post with a new showcase image so recommendation accounts have fresh material to repost every week. When Chinese interest fades, publish an English comparison and a multilingual showcase page. Foreign repost accounts will give you a second wave, which is how oil-ui's October peak happened.

**When to move** : There's no deadline, but the window is shrinking. Agent skill directories grow every week, and big players occupy each cell quickly. The earlier you pick an empty cell, the better.

**What would show you misread this** : ① A vertical design skill from an author with no following reaching a thousand stars in a month would mean this category can take off on its topic alone, not just on author fame. ② If oil-ui keeps gaining more than 50 stars a day after its author stops posting, its workflow is spreading by word of mouth. ③ Conversely, if its skills.sh installs stay in the low thousands, most of those stars are spectators. People building similar tools should decide whether they want stars or real users.

[GitHub repo ↗](https://github.com/oil-oil/oil-ui)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| pbakaus/impeccable | Open source | 2025-11-16 | 103,756 stars; skills.sh 1.9M installs | A set of design commands for polishing and review; oil-ui does exploration from scratch across multiple directions plus comparison |
| nextlevelbuilder/ui-ux-pro-max-skill | Open source | 2025-11-30 | 145,640 stars; 913K installs | A library of style, color and typography rules; oil-ui is a methodology plus comparison workflow |
| Anthropic frontend-design(anthropics/skills) | Big-company features | 2025-09 | 187,695 stars; 3.7M installs | Official general aesthetics guide, single option, no comparison pages or blind screenshot review |
| Leonxlnx/taste-skill | Open source | 2026-02-19 | 111,146 star | An aesthetic rules pack with a single direction |
| Nutlope/hallmark | Open source | 2026-04-27 | 48,631 stars; 94K installs | Anti-AI-slop rules, big-account author brings traffic |
| alchaincyf/huashu-design | Open source | 2026-04-19 | 28,887 star | A multi-style visual design skill from a Chinese author, the closest Chinese competitor |
| emilkowalski/skills | Open source | 2026-03-16 | 54,085 star | Design engineering principles for motion and details |
| Nutlope/variate | Open source | 2026-07 (approx.) | 221 star | Generates 4 variants of the same component, component level only |
| heydamir/unslop | Open source | 2026-10-29 | 9,647 stars; 1.3K installs | Appeared a month later, specifically turns AI-slop pages into ones that look designed, and grew faster |
| feimacode/open-design-agent-kit | Open source | 2026-09-19 | 1,028 star | A large suite of 270+ design skills |

Where users come from

- **Verified** **Author's own X and Bilibili launch** : 9-30 X launch post 12.6k views; same-day Bilibili video about 5,757 plays; 10-03 and 10-06 update posts each about 66k views

- **Verified** **Reposts by Chinese tech accounts** : Programmer Yupi 10-04 (17.8k views), GitHubDaily 10-06, GitTrend 10-23, awe0 11-04; Sumi Ke, AI Partner, Zhihu reposts

- **Verified** **Foreign repost and design accounts** : gittrend.io Instagram Reel 10-05; @Yarad_ai in Persian 10-25 (16k views, right next to 10-26/27's +138/+135); @dopamine_crypto in Korean 11-04 (16k views); @101babich 11-05; several LinkedIn posts; explainx and Setproduct articles

- **Verified** **Skill directory listings** : skills.sh, Cola Skill (the author announced the listing 10-09), mcpmarket, skillsllm, typingmind, ui-skills and others

- **Verified** **Multilingual SEO and showcase site** : ui.oiloil.org's sitemap has 870 URLs in 6 languages, including showcase pages, guide pages, and comparison pages against big players like Impeccable; README links carry utm tags

- **Verified** **New platform plugin** : 10-18 built dsh-oil-ui, a native DeepSeek Harness plugin (214 stars)

- Inferred **Hacker News and paid ads** : Not found in the hackernews database; no sign of paid promotion

## Why oil-oil/oil-ui took off

### Bottom line

oil-ui 能起来,靠的是一个在中文 AI 圈已经有观众的独立开发者。他有 X 约 1 万粉,B 站「oil欧呦」2.08 万粉。他把自己讲过的「让 AI 做出不平庸设计」的方法论做成 skill,9-30 在 X 和 B 站同日发布;第 2–7 天被程序员鱼皮、GitHubDaily、gittrend.io 等中文和英文搬运号自发转发,第一周就破了 1,000 star。此后他几乎每天发版,每次更新都配新作品图在 X 上发帖;再加上 870 页、6 种语言的作品画廊和 SEO 站,以及韩语、波斯语、英文设计号的二次搬运,涨星一直维持在日均 30–140。

Its differentiator is 'generate several genuinely different directions side by side, then polish against real screenshots'. That differs from the big players that feed rule libraries to the AI (Impeccable, UI UX Pro Max), and it's the line reposters most often summarize. But differentiation alone didn't make it take off: the big players in the same space have stars and installs two orders of magnitude higher. The main reason it grew is the author's built-in traffic plus high-frequency shipping and public posting.

### Key facts

- 建库 2026-09-30,MIT。截至 2026-11-06,2,805 star、156 fork(我们的 github 库);GitHub API 现查是 2,849 star、162 fork。

- Star growth curve (github database, repo_day): 10-09 with 1,085 → 10-17 with 1,585 → 10-24 with 2,058 → 11-06 with 2,805. Single-day peak during the observation period was +138 (10-26) and +135 (10-27). Over the last 7 days it added 309, versus +481 over the prior 7 days. It's cooling, though it rose again to +76 on 11-05.

- 早期:作者 10-06 在 X 上说「两天有 570 star」(https://x.com/I_am_oil_oil/status/2107146592797258061 )。他 10-03 的帖子写的是「这两天开源了」,所以仓库可能是 10-03 前后才真正公开推广,或者 9-30 发布后又集中推了一轮(未核实)。10-07 发帖时「快 1k star」(https://x.com/I_am_oil_oil/status/2107847848863613315 )。

- Real usage: skills.sh shows oil-ui with 3.3K total installs (2026-11-06 web). On the same page, Impeccable has 1.9M, UI UX Pro Max 913K, Hallmark 94K, Anthropic skills 3.7M. Stars and installs are far apart, so few people are actually using it.

- Releases: About 37 days to v0.32.0 (GitHub releases and commit history), with 10-11 shipping 5 versions in one day.

- It's not the first version: the author 2026-04-24 built draw-ui (a UI design skill, 544 stars), and oil-ui is the second-generation rebuild.

### The author and his playbook

- Author Zhihuang Lin (GitHub oil-oil, 1,634 followers, account registered 2019 years ago). X @I_am_oil_oil has about 10,055 followers (fxtwitter API), and Bilibili 'oil欧呦' has 20,755 followers (bilibili relation API, mid 13030183). His main work is talking-head videos and posts about AI tools, with daily video uploads.

- The same author has 30 skill and tool repos, about 13k stars combined: oil-motion 2.8k, beautify-github-readme 1.95k, wolfcha 746, draw-ui 544, and more. Each repo's README cross-promotes the others (the oil-ui README ends by pushing oil-motion and oil-tone).

### Growth tactics (evidence)

1. **Author's own content launch (confirmed)** . 2026-09-30 15:54 X launch post: it says this skill is the methodology from an earlier video, and also promotes the paid Oil UI Pro. The post has 167 likes and 12.6k views (https://x.com/I_am_oil_oil/status/2105204499266855051). The same day, 15:52, Bilibili got a 7-minute 48-second video titled 'This Skill Helps You Make Better-Looking UI Designs!', with about 5,757 plays (Bilibili page, BV158af6zEY9).

2. **Author's running 'new showcase + new release' posts (confirmed)** .
    - 10-03 post introducing the showcase gallery website: 492 likes, 66k views (https://x.com/I_am_oil_oil/status/2106199616039587934).
    - 10-06 post on 'design capability keeps evolving': 463 likes, 67k views (https://x.com/I_am_oil_oil/status/2107390277342400666).
    - 10-14 post with a Gemini 4 Pro review, also promoting Oil UI Pro: 100k views (https://x.com/I_am_oil_oil/status/2110047189963788656).
    - Other updates include 10-24's '2k stars' post, 11-01's '2600 stars + admin-panel capabilities' post, and 11-03's design-spec governance post.

3. **Organic reposts by Chinese tech accounts (confirmed)** .
    - Programmer Yupi 10-04 post: 156 likes, 17.8k views (https://x.com/yupi996/status/2106618315779252596).
    - GitHubDaily 10-06 post: 60 likes, 6,913 views (https://x.com/GitHub_Daily/status/2107372841180311578).
    - GitTrend 10-23 post (https://x.com/GitTrend0x/status/2113593806834585978).
    - awe0 11-04's post '6 open-source projects that add taste to AI' lists it alongside UI UX Pro Max and Huashu Design: 143 likes (https://x.com/awe0_io/status/2117847923883209110).
    - Reposted on Chinese sites: Sumikeq (10-07), Erqu (10-14), AI partner "2026 most popular 4 open-source frontend UI Skills" (10-31), Zhihu column.

4. **Second wave from foreign repost accounts (confirmed; corresponds to the plateau after 10-25 and the rebound)** .
    - English: gittrend.io posted an Instagram Reel on 10-05 (https://www.instagram.com/reel/DeHs15riSZr/ ); The Daily Commit 10-06; LinkedIn posts 10-11, 10-20 (Portuguese), 10-30; explainx.ai ran an article titled "1.6K Stars in 2 Weeks" (10-17); novidadesia.com 10-13 (Portuguese); Setproduct 11-03, "13 design skills for AI agents compared"; Nick Babich (@101babich) 11-05 put it first in a post: 194 likes, 11.7k views (https://x.com/101babich/status/2118342559088910438 ).
    - Persian: @Yarad_ai 10-25 "Oil UI breaks 2 stars", 324 likes, 16k views (https://x.com/Yarad_ai/status/2114024642897723845 ), posted right next to the +138/+135 peak on 10-26/27.
    - Korean: @dopamine_crypto 11-04 "A Claude skill for UI design that hit 2,208 stars in a month", 398 likes, 16k views (https://x.com/dopamine_crypto/status/2117749915602583707 ); stars rose +76 after 11-05.

5. **Seeding skill directories (confirmed)** : skills.sh, Cola Skill (the author posted the listing announcement, 35.6k views, by 10-09), mcpmarket, skillsllm, typingmind, ui-skills.com, claude-updates, skillsfordesign, trendshift, tool.lu and others have listed it.

6. **Multilingual SEO site (confirmed, effect unknown)** : the ui.oiloil.org sitemap has 870 URLs covering 6 languages: Chinese, English, Japanese, Korean, German and Spanish. Each language has 117 showcase pages, 14 guide pages and 11 "inspiration" category pages, e.g. /en/guide/claude-code-ui-design/ and /en/guide/fix-ai-ui-slop/. There is also a comparison piece, "4 AI UI Design Skills Tested Side by Side", that puts it against Anthropic frontend-design, UI UX Pro Max and Impeccable, riding their search traffic. Every README link carries utm tags, suggesting he is measuring GitHub-to-site conversion.

7. **Claiming new platforms (confirmed)** : 10-18 built a native plugin, dsh-oil-ui (214 stars), specifically for DeepSeek Harness. A search snippet shows a Bilibili video, "DSH Oil UI plugin is here", at about 87k plays (seen only in the search summary, not verified).

8. **Not found** : it is not on HN (the hackernews library returns nothing); no sign of paid ads; GitHub trending only covers repos under 60 days old, so it never charted, which tells us nothing either way.

### Monetization (not verified, only what we saw)

- When 9-30 launched, the Pro version sat behind Alipay SkillPay on a subscription model. Users couldn't install it, and 9 people got refunds. The author 10-03 posted a long public post-mortem (https://x.com/I_am_oil_oil/status/2106266121855005072 ), and AGI Hunt wrote a report on it. The setup was then rebuilt in-house: a one-time $9.99 purchase, Stripe checkout (supporting Alipay, WeChat Pay and Link), and a self-written oil-cli that handles device-code login, pulls the Pro version from a private repo after payment is verified, and prompts for updates.

- 10-07 posted the full technical stack and costs publicly (domain about $10, Resend $20 a month), and funneled readers to oil-skill-creator (https://x.com/I_am_oil_oil/status/2107847848863613315 ). The open-source edition is generated from the Pro version by a script that strips it down, and both versions ship together.

- No sales or revenue have been published, and there is no TrustMRR record. Posts by 10-17 and 10-31 mention "Pro users" and a case study (a Hong Kong research institute redid its website with Pro), so people are clearly buying, but the scale is unknown.

### Competitors

| Name | Where | Launched | Scale | How oil-ui differs |
|---|---|---|---|---|
| pbakaus/impeccable | Open source | 2025-11-16 | 103,756 stars, +8,787 in 7 days; skills.sh 1.9M installs | a set of design verb commands (polish, bolder, critique) for polishing and reviewing existing interfaces; oil-ui generates multiple directions from scratch and then picks among them |
| nextlevelbuilder/ui-ux-pro-max-skill | Open source | 2025-11-30 | 145,640 stars; 913K installs | a large rule library for style, color and type; oil-ui is a methodology plus a comparison workflow |
| Anthropic frontend-design(anthropics/skills) | Big-company features | From ¥2025-09 | 187,695 stars in the repo; 3.7M installs | an official general-purpose frontend aesthetics guide that works out of the box; no multi-direction comparison or screenshot-based review |
| Leonxlnx/taste-skill | Open source | 2026-02-19 | 111,146 star | a "taste" rule pack, leaning toward a single high-end aesthetic direction |
| Nutlope/hallmark | Open source | 2026-04-27 | 48,631 stars; 94K installs | anti-"AI look" rules; the author brings a large following of his own |
| emilkowalski/skills | Open source | 2026-03-16 | 54,085 star | Design engineering principles for motion and details |
| alchaincyf/huashu-design | Open source | 2026-04-19 | 28,887 star | a Chinese creator doing multi-style visual design, the closest Chinese competitor |
| Nutlope/variate | Open source | About 2026-07 | 221 star | the same component rendered in 4 variants; a predecessor of the "side-by-side options" idea, but only at the component level |
| heydamir/unslop | Open source | 2026-10-29 | 9,647 stars; 1.3K installs | launched a month later, turning vibecoded pages into "designed" ones; growing faster than it did |
| feimacode/open-design-agent-kit | Open source | 2026-09-19 | 1,028 stars, 7 days +589 | a large suite of 270+ design skills, launched around the same time |
| oil-oil/draw-ui | open source (same author) | 2026-04-24 | 544 star | the author's previous generation, which produced design mockups and then rebuilt them |

### Why it took off

1. **Built-in distribution from the author** : 10k followers on X and 20k on Bilibili, so there were videos and posts the day it launched. oil-ui was not a cold start; the author's earlier two repos, oil-motion and beautify-github-readme, grew the same way.

2. **Hits a pain everyone shares** : "AI-built pages look AI-built at a glance", with purple gradients, centered big headlines and three rounded cards. Nearly every repost opens with this line.

3. **A difference you can state in one sentence** : "generate several truly different directions side by side, then refine against real screenshots". Repost accounts find this easy to repeat (the Korean post, Nick Babich and awe0 all summarize it this way).

4. **Showcase gallery** : every piece comes with its prompt, labeled with the cheaper model used (DeepSeek, GPT Luna). This answers the demand for making cheap models produce good design, and gives repost accounts ready-made images.

5. **High-frequency iteration, with a post for every release** : 37 minor versions in 32 days, with new content on X every week, forming a steady series of small peaks.

6. **Publicly reviewing the failure and tech stack of a paid skill** : these two posts are shareable in their own right (AGI Hunt covered them), and they tie the question of whether paid skills can sell to this project.

### Differentiation

- The mainstream UI skills (Impeccable, UI UX Pro Max, taste, Hallmark, frontend-design) mostly do the same thing: feed the model aesthetic rules and an anti-AI-look checklist, and output a single answer. oil-ui differs in its workflow. It first breaks down the category and benchmark products, translates "premium" and "clean" into concrete design decisions, generates several sharply different directions, lines them up on a built-in comparison page to choose from, runs blind reviews at desktop and mobile sizes with screenshots, and then subtracts. It also has a dual open-source and paid structure, where the open-source edition is essentially a trial version of Pro.

- The differentiation helped spread it, but it was not the main driver of growth. Similar projects from the same period (Hallmark, unslop, taste) got 3–40 times as many stars, on the strength of the author's reputation or a single "anti-AI look" hook. oil-ui's 2.8k stars and 3.3K installs match the circle the author's personal influence can reach.

### What to borrow

- The combination of "methodology video → skill → showcase gallery → paid Pro": productizing content the author had already presented, with videos and posts as the launch channel.

- Only the Pro version is maintained; the open-source edition is generated by trimming it with a script. Payment, install and updates run through a CLI with device-code login. The barrier is not the source code but the ongoing updates.

- The showcase page attaches prompts and models, alongside an SEO site in 6 languages, plus a comparison article against the big players to capture search terms.

- A new agent platform (DeepSeek Harness) got a native plugin right away, to claim the spot early.

### Unknowns

- We have no day-by-day star data for 9-30 to 10-08, and which post drove the first-week surge is unverified. Which exposure the 10-26/27 peak corresponds to is only a guess, possibly tied to Yarad's post.

- Pro sales and revenue are not public.

- Traffic from the SEO site is unknown.

- Whether the "87k plays" Bilibili video was posted by the author himself and whether the play count is accurate is unverified.

### Sources

- Our github library: repo_profile / repo_day (2026-11-06); GitHub API (2026-11-06)

- X library live lookup (TikHub), fxtwitter API: the author's and reposted posts, links above

- skills.sh repo pages (2026-11-06); ui.oiloil.org/sitemap.xml, /en/pro/, /en/guide/ui-design-skills-tested/

- Bilibili pages and the relation API; reposts across sites found via Exa and Jina search (explainx.ai, setproduct.com, agihunt.info, xmsumi.com, ipd.me, etc.)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.