# "301 Opus 5.5 videos, and I spent $3,000 on them": how a prompt collection got 14k stars in the hottest 72 hours · GitHub open-source project

> Source: https://stringers.io/en/github/gh_yihui-dev__awesome-opus5-5-videos.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

> Opus 5.5 coding and motion-video demos took over X, everyone was asking for the prompts, and it dropped on the day demand peaked: huge volume, results you could see, and the creator's main account launched it, then English AI repost accounts and blogs amplified it in a pile-on.

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

yihui-dev/awesome-opus5-5-videos · yihui-dev

# "301 Opus 5.5 videos, and I spent $3,000 on them": how a prompt collection got 14k stars in 72 hours at its peak

**What it is** A GitHub list of 703 prompts for the Claude Opus 5.5 viral code-animation videos. The homepage is a wall of GIF previews, and 418 of the entries can be compared against the originals and remakes on the author's own site.

**Why it blew up** Opus 5.5 writing code that produces animated videos was flooding X, and everyone was asking for the prompts. The repo launched on the peak demand day, the results were big and easy to see, and after the author's large account posted it first, English AI repost accounts and blogs amplified it.

**C** **How you could follow it** Don't build another Opus collection. Build a pipeline in advance that scrapes prompts and auto-renders previews, so you can ship within 48 hours when the next model produces a viral visual moment. Also fix what this one left undone: whether a prompt is safe to paste and whether it reproduces as-is. Or copy the runner-up and turn a style that reliably produces good videos into an installable skill.

Grade C: you can learn a full playbook for earning stars here, but this wave is past its peak and there are over a dozen similar lists. Its success came from a one-off visual moment plus the author's own big account. Using the same template on four other models, the author got only 14 to 87 stars, so followers would struggle to replicate it. Publicly known / widely discussed

Data (updated daily) · Repo created 2026-09-27 · 14,047 stars · 7 days +1,145

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

On Sep 2026 27, at 3:54 UTC, a GitHub account called yihui-dev pushed its first commit: a list of 282 prompts, one for each viral video made with Claude Opus 5.5 writing the code. Over 4 hours later, the author Yihui (@yihui_indie, about 64.5k followers) posted a Chinese-language thread on X: "🔥301 Opus 5.5 videos, and I spent a full $3,000 on them. I used Claude Opus 5.5 to rebuild the hardest, most viral effects from across the internet... All the test videos, the full prompts and the workflow, organized and open-sourced!" In the first reply he put the repo link, with "Would love a Star ⭐️" in parentheses. The post later reached 99.8k views. Less than an hour after that, Elon Musk reposted an Opus 5.5 video by @anabology with just the word "Wow", getting 14k likes; that video was the first entry in the repo's "motion" category. By Nov 6, the repo had 13,875 stars and 1,465 forks (GitHub, 11-06).

### What it is

It's an awesome-list. The first line of the README reads "Viral videos people made with Claude Opus 5.5, each with the prompt the creator shared." Everything in it is the same kind of thing: getting Opus 5.5 to write animation directly as code (HTML, Canvas, SVG, Three.js, GSAP), where the rendered output is a video. The repo now holds 703 prompts, each in its own Markdown file, tagged by category (motion, explainer, 3D, game), tech used, and original creator, with a link back to the creator's X post.

Compared with other lists from the same period, it had three extra things:

- **You can see the results right away.** The README homepage is a GIF wall built from 100 of the entries. You know what each prompt produces without clicking through, which makes it easy to screenshot and share.

- **It was actually reproduced.** The author said he spent about $3,000 re-running several hundred of the entries. 418 of them now appear on his own site, Skillry, side by side with the originals and remakes, answering the question readers care about most: if I take this prompt, will it actually work?

- **Biggest volume.** It launched with 282 entries, and two days later it had 475. 10-12 once added 335 entries, bringing the total to 853. Then 10-13 author removed 150 entries, saying they were "not made purely with Opus 5.5 prompts," leaving 703. There have been no updates since.

The repo is also the author's traffic funnel. The large image in the README, the "Watch all 418 side by side" button, and every GIF carry tracking parameters pointing to the Opus 5.5 video page on skillry.dev. That page lets you copy the prompts for free, and scrolling down shows the video and slide skills he sells.

### Why this one: the timing was too perfect

Look at how the demand started. Opus 5.5 was released by 9-22. On 9-24, @shneural sent the same prompt, "make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are...", to Opus 5.5 and GPT 6 Astra, then posted the two outputs side by side. That post got 1.12M views. The next day @ajith_io ran the same prompt again for 186k views, and a batch of developer accounts followed by showing off their own Opus videos. TikTok did the same: a Spanish-speaking creator said "follow me and I'll post the prompts" and got 1.46M plays, and another video, "How to do motion design with Claude Opus 5.5", got 147k plays.

Every comment section was asking the same question: "What's the prompt for this video?" That question peaked around 9-27, which is exactly when this repo appeared.

There were other entries in the same window. athemeroy's list was created 9-25, two days earlier. joeseesun's Chinese version was created the same day. But the former had no preview wall and no big-account author, and the latter had only 18 prompts. The one that became the default reference for "that Opus video collection" was the one with the most volume, GIFs, remakes, and a first-posting author with over 60k followers.

### Where the stars came from

| Timeline | stars | What happened |
|---|---|---|
| 9-27 | Repo created | The author's Chinese X post asked for stars right away; most accounts that forked it that day had Chinese usernames. |
| 9-28 to 10-01 | — | The first wave of English aggregator blogs reposted it, with titles like "475 prompts…" that lead with a number. |
| About 10-09 | About 3,500 | A LinkedIn author with 20k followers posted that it had "just crossed 3,500 stars." |
| 10-17 | 5,439 |  |
| 10-21 → 10-22 | 6,962 → 8,115 | +1,153 in a single day, the peak for the whole run. |
| 10-28 | 12,104 | Single day +670 |
| 11-06 | 13,875 | Last 7 days +1,203, the 7 days before that +3,462. It's cooling. |

(Star counts: GitHub daily figures. The 10-09 figure comes from the LinkedIn post.)

It breaks into three phases:

**Phase one was the author's own account.** The Chinese post went out first, and the comments asked directly for stars. The first stars came from the Chinese AI community.

**Phase two was English repost accounts, and it was the largest phase.** From 10-21 to 10-24, @simplifyinai on Threads, Ram Maheshwari on LinkedIn (opening with "WOW.... this is an insane collection…"), a Facebook page copying the same copy, an Instagram reel trading engagement for "comment Viral for the link", plus blogs like agentgid, Chen Media, and daily.dev, all reposting within a few days. That matches the +1,153 on 10-22. From the outside, it's impossible to tell whether these accounts reposted organically or were paid (unverified). Judging by their style, this kind of account favors repos with a big number, free access, and GIFs, and their titles nearly always follow the pattern "703 Prompts & 418 Remakes".

**Phase three was the long tail.** German, Thai, and Spanish blogs plus a pile of GitHub trending aggregators kept reposting into early November.

One caveat: the stargazer list endpoint for this repo wouldn't load, so I couldn't check the accounts that starred it one by one. About 10.6% of the stars come from forks and there are only 54 watchers. That fits a bookmark-style list, where many people fork a copy to keep, but star-botting can't be fully ruled out.

### Same template, different model, and it stops working

This is the most telling part of the whole story. Between 10-05 and 10-24, the author used the exact same README, the same GIF wall, and the same remake page to build four more collections: Fable 5.5, Gemini 4 Argon, Kimi K3.1, and Claude Motion (GitHub, 11-06):

| List | stars |
|---|---|
| awesome-opus5-5-videos (this one) | 13,875 |
| awesome-fable5-5-videos | 87 |
| Gemini 4 Argon edition | 33 |
| Kimi K3.1 edition | 29 |
| Claude Motion edition | 14 |

Same method, same author, same big account, and the results differ by two orders of magnitude. One cycle earlier, when GPT-6 Astra came out, a batch of prompt collections with the same pattern appeared, and the biggest reached 121 stars.

The conclusion is clear: copying the template can't reproduce the result. What decides it is whether the model produces a visual moment that floods the internet and makes everyone want to copy it. Opus 5.5's "write an animated video from one sentence" was that moment. Fable 5.5 and Kimi didn't have one.

### Who it's competing with

| Project | Repo created | Stars (11-06) | How it differs |
|---|---|---|---|
| lemomo-ai/lemo-opuscar | 9-26 | 12,987 | A Claude Code plugin that bundles 66 video styles one person tuned, so you can install it and generate videos right away. It was second-most starred in the same wave. |
| Vincentwei1021/mg-styles-15 | 10-02 | 3,190 | 15 styles the author built himself, each with a full prompt and source code. It's not a collection. |
| athemeroy/awesome-claude-5-5-videos | 9-25 | 963 | Created two days earlier. It includes source verification and covers several Claude models. No preview wall and no remakes. |
| joeseesun/opus-video-prompts | 9-27 | 968 | In Chinese, with 18 prompts and 54 examples. |
| zhuyansen/awesome-opus-5.5-video | 9-27 | 307 | Similar lists created the same day. |
| opusvideo/awesome-claude-video | 9-25 | 238 | A similar list, bilingual Chinese and English. |
| HiAPIAI/awesome-opus-5-5-video-styles | 9-29 | 54 | Regroups the 425 prompts into 12 visual styles. |

Copycats have already appeared: motionvideohq's Fable 5.5 motion collection, with descriptions nearly identical word for word; mailes combined Opus and Fable into "587 viral videos"; plus a 3D-focused edition, a general "AI writes code to make videos" roundup, and a web version called fable5.app.

The runner-up, lemo-opuscar, is worth noting. It doesn't collect other people's prompts. It tuned its own styles, packaged them as a tool, and still reached 13k stars. In the same wave, both "curating others' work" and "building something usable yourself" earned stars, and the second can't be copied as easily.

### Where it falls short

- **Pasting these prompts isn't necessarily safe.** 10-27 Someone opened a long issue naming 5 prompts. Some tell the agent to clone a repo hidden behind a short link and "override default settings" with the rules in it. Some contain the original author's own folder paths and a paid API key, and tell the agent to "use up the quota." Some directly modify the agent's global config. If someone pastes them unchanged into an agent with tools, it could run unfamiliar code and spend their money. As of 11-06, the issue had no reply.

- **The "pure prompt" rule wasn't enforced all the way.** The author later tightened the standard (10-13 removed 150 entries with it), accepting only prompts "written directly to Opus 5.5 that can be reproduced by pasting." He used this to reject 6 videos submitted by others. But the top entry, from @anabology, asks for image-generation and video-generation API calls, ElevenLabs for sound effects, and reading the author's local asset folder. An ordinary person can't reproduce it by pasting.

- **Updates stopped after 10-13.** Stars kept climbing, but the content hadn't changed in over a month. Most issues opened after mid-October (some submissions, some asking "does it work with Codex?") went unanswered.

Those are the places a follower could step in.

### If it were you

**In short:** this list was the result of three things coming together: a new model, a visual moment, and a big account posting first. Opus 5.5 has already passed its peak in this wave, and similar lists are crowded. If you want to learn from it, learn to ship the biggest, best-looking collection within 48 hours when the next moment arrives. Whether that works depends mostly on luck and on whether you have an audience.

**Who it's for:** developers who already have tens of thousands of AI followers on X, Jike, or Xiaohongshu and get on every model launch first; people willing to spend several hundred to several thousand dollars of model credits on remakes and sort through hundreds of entries a day for several days.

**Who it's not for:** anyone without their own distribution channel. The author used the same template on four models and only got double-digit stars. It's also not for anyone thinking of building "yet another Opus 5.5 video collection" now, since that slot already has over a dozen entries.

**How to do it:** don't copy the Opus version. Prepare for the next one in advance: build a pipeline that goes "scrape posts → extract prompts → auto-render preview GIFs → generate the README". When a new model launches, watch X for posts where the same prompt is being shared with results, and ship within 48 hours. Fix what this one missed at launch: tag each prompt with "is it safe to paste" and "does it need extra APIs or assets", expand short links, remove anything that only exists on the original creator's machine, and make a version that truly runs when copied. The other route is to copy the runner-up: don't collect, tune a few styles that reliably produce good videos, and package them as a skill you can install directly. That's harder to copy.

**How to get users:** Launch on your own account first and ask for stars outright; in the post, include the most striking finished clip and one big number ("301 videos", "$3,000"). Once it's live, hand the GIF wall and numbers to English AI repost accounts and GitHub trending aggregators. They love "big number + free + has animated GIFs." Credit every item to the original creator and link back, and the original creator might repost it too.

**When it was before:** This Opus 5.5 wave was +7 over the first 3,462 days and +7 over the last 1,203 days, clearly cooling. Getting in now is too late. The next window depends on the next visual peak after a model launch, and its timing can't be predicted. When it comes, you only get a window of a few days after launch (inferred).

**How you'd misread it:** ① After the next flagship model launches, no "same prompt shared across X" posts show up within a week, which means this kind of peak was a one-off for Opus 5.5; ② you launched within 48 hours of the peak, and your own launch post brought in fewer than 500 stars; ③ the model vendor or a site like Skillry publishes its own official "prompts + remakes" gallery, so the list has no reason to be bookmarked.

[GitHub repo ↗](https://github.com/yihui-dev/awesome-opus5-5-videos)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| lemomo-ai/lemo-opuscar | Open source | 2026-09-26 | 12,987 stars | It's an installable Claude Code plugin, with one person's tuned video style that can output finished clips directly. It collects other people's prompts. |
| athemeroy/awesome-claude-5-5-videos | Open source | 2026-09-25 | 963 stars | Two days earlier, leaning toward research notes and source verification. No GIF wall, no remakes, and the author has no big account. |
| joeseesun/opus-video-prompts | Open source | 2026-09-27 | 968 stars | In Chinese, with only 18 prompts and 54 case studies, and no license. |
| Vincentwei1021/mg-styles-15 | Open source | 2026-10-02 | 3,190 stars | The author made 15 styles themselves, with source code. It's not a collection. |
| zhuyansen/awesome-opus-5.5-video | Open source | 2026-09-27 | 307 stars | A same-category list built the same day, but smaller in volume and distribution than this one. |
| opusvideo/awesome-claude-video | Open source | 2026-09-25 | 238 stars | A same-category list, bilingual Chinese and English, with no remakes. |
| HiAPIAI/awesome-opus-5-5-video-styles | Open source | 2026-09-29 | 54 stars | Re-sorts 425 prompts into 12 visual styles. |
| fable5.app/opus-5-5/works | Web | Not verified | 682 videos | A web-page collection with no GitHub star backing. |
| heygen-com/hyperframes | Open source | 2026-03-10 | 80,024 stars | It's a framework that renders HTML into video, not a prompt library. |

Where users come from

- **Verified** **Platform event: Opus 5.5 video takes over the feed** : 9-22 Opus 5.5 launched; 9-24 @shneural posted a same-prompt comparison, 1.12M views; 9-27 Musk reposted @anabology's video with "Wow", 14k likes; the repo was created at 03:54 UTC the same day, and @anabology was the first entry in it (x library).

- **Verified** **Author's big Chinese account launches it and asks for stars** : @yihui_indie posted on 9-27 08:16: "301 Opus 5.5 videos, I spent a full $3,000...", 99.8K views, comments reading "GitHub open source repo (asking for a Star ⭐️)"; the author has 64.5k followers (TrustMRR); 9-27 most of the day's forks came from accounts with Chinese usernames.

- **Verified** **English AI repost accounts (LinkedIn / Threads / Facebook / Instagram)** : LinkedIn James Chang (28k followers), Arthi R. (10-09), Ram Maheshwari "WOW.... insane collection" (10-22); Threads @simplifyinai (10-21); a Facebook page copying the same copy; an Instagram reel reading "Comment Viral for link"; the timing matches the +10-22 star peak of 1,153.

- **Verified** **SEO / aggregator blogs, multilingual** : frontiernow 9-28, opensourceprojects.dev 9-30, AGI Hunt 10-01, thanisch.co (German) 10-15, Chen Media 10-21, agentcookbook / agentgid 10-22, aiidelist / daily.dev 10-24, ailistscout 10-27, vibecodingthailand (Thai) 10-31, charliehills 11-03, codeia (Spanish).

- **Verified** **Author pays for remake content + on-site SEO detail pages on Skillry** : self-reported spend of about $3,000 to remake 301 videos; skillry.dev/ai-videos/opus-5-5 has one detail page per video; every README link carries UTM parameters pointing to Skillry.

- Inferred **Paid promotion / star farming** : no evidence found; the stargazers endpoint returned 404, so the quality of starring accounts can't be checked.

## yihui-dev/awesome-opus5-5-videos: why it took off

> Channel: GitHub | repo created 2026-09-27 | 11-06 already has 13,871 stars and 1,466 forks (GitHub API) | author Yihui / 熠辉 (@yihui_indie), who runs Skillry (skillry.dev)

### Bottom line

It's a **traffic-funneling list** from the Opus 5.5 video craze. It reached 14k stars on three things:

1. **Timing** : Opus 5.5 9-22 launched, 9-24 to 9-27 "coding with Opus to make motion-design videos" went viral on X, and Musk 9-27 even reposted one; the author started the library on the same day as this window.

2. **Scale** : It launched with 282 prompts, while other collections at the time had only 18 to 50. The author also claimed to have spent about $3,000 recreating 301 videos, and puts originals and remakes side by side on Skillry.

3. **Distribution** : The launch came from the author's Chinese X account with 64.5k followers, the post got 99.8K views, and comments read "asking for a Star" directly. English AI repost accounts (LinkedIn, Threads, Facebook, Instagram) and a wave of SEO blogs then picked it up.

The library itself doesn't make money. It funnels traffic into Skillry's paid skill subscriptions. Skillry's revenue is public on TrustMRR: September $1,608, October $12,216, MRR $4,647,434 from active subscriptions.

The same author used the same template for four other collections, Fable 5.5, Gemini 4 Argon, Kimi K3.1, and Claude Motion, and each got only 14 to 87 stars. That shows the thing that worked was the window when everyone was searching for Opus video prompts, not the template, which can't be copied to get the same result.

---

### 1. What it is

- The README title is "Viral videos people made with Claude Opus 5.5, each with the prompt the creator shared". It currently collects 703 prompts (`prompts/*.md` and `data/videos.json`), and 418 of them have a "live remake" on Skillry. The README's front page picks 100 items for the GIF wall, in four categories: motion, explainers, 3D, and games.

- Every item links back to the original creator's X post and credits them by name (@anabology, @moritzkremb, @melvynx, @daniel_mac8...).

- The large banner at the top of the README, the "Watch all 418 side by side" link, and the Browse link at the end all carry `utm_campaign=awesome-opus5-5-videos` and point to skillry.dev/ai-videos/opus-5-5.

- Skillry's page lets you view originals and remakes side by side and copy prompts for free. The lower half of the page is the "Unlock every Premium Skill" pricing section: $9.99/month, $79/year, $169 lifetime, currently marked "Black Friday 50% OFF", and it says "1,000+ people supporting us".

- The commit history (GitHub API) records how the repo grew:
    - 9-27 03:54 UTC first commit, 282 entries
    - 9-28 grew to 389 entries, adding a mosaic hero image and Skillry screenshots
    - 9-29 grew to 475 entries
    - 10-08 grew to 513 entries
    - 10-12 added 335 entries at once, reaching 853
    - 10-13 removed 150 entries that were "not made with pure Opus 5.5 prompts" (HyperFrames demos, outputs from other skills), leaving 703
    - 10-13 no pushes after that

### 2. Growth data

| Timeline | Stars | Source |
|---|---|---|
| 9-27 | Repo created | GitHub API |
| About 10-09 | About 3,500 | LinkedIn Arthi R. 10-09 post "just crossed 3,500 stars" |
| 10-17 | 5,439 | Data card daily |
| 10-21 → 10-22 | 6,962 → 8,115 (1,153 in a single day, the peak) | Data card |
| 10-22 | 8,978 (API value that day) | agentgid.com 10-22 article |
| 10-28 | 12,104 | Data card / GitHub library |
| 11-06 | 13,737 (card) / 13,871 (live API) | Data card / GitHub API |

- Gained 7 stars in the last 1,203 days, and 7 stars in the first 3,462 days, clearly cooling.

- 1,466 forks, a fork-to-star ratio of about 10.6%, which is high for a "bookmark-type" awesome list (many people fork it to keep a copy). Only 54 watchers.

- The GitHub stargazers endpoint returned 404 for this repo, so **the quality of the starring accounts can't be checked one by one** . Many early fork accounts had Chinese usernames (9-27 that day: JRS-Space, shaomingchan, xinyiheng, DamonQ123, yinwenjie...), which lines up with the timing of the Chinese launch post.

Skillry's revenue moved in step with the repo's star growth (TrustMRR, verified via the Stripe API):

| Month | Skillry revenue |
|---|---|
| 2026-08 | $27 |
| 2026-09 | $1,608 |
| 2026-10 | $12,216 |
| 2026-11 (through day 6) | $1,894 |

- $15,745 lifetime, $4,647,434 current MRR from active subscriptions, solo team, TrustMRR lists the country as China.

- The two daily revenue peaks, $10-21 on 901 and $10-28 on 870, line up with the English repost waves (10-21 / 10-22) and the day 10-28 gained +670 stars in one day.

### 3. Why it took off

#### 3.1 Platform event: Opus 5.5 video takeover (timing)

- 9-22 Anthropic launches Claude Opus 5.5 (@claudeai, x library).

- 9-24 @shneural ran the same prompt, "make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are...", against Opus 5.5 and GPT 6 Astra. The post got 1,119,942 views and 5,469 likes (x library).

- 9-25 @ajith_io ran the same prompt and got 185,618 views; @stephanlivera (190k followers), @DavidKPiano, @deedydas and others were posting Opus 5.5 videos (x library).

- 9-27 09:04 UTC, Musk reposted @anabology's Opus 5.5 video with "Wow", 14,111 likes (x library). That post is the first entry under Motion graphics in the repo.

- There was clear demand on TikTok too: @itskeo5 "Síganme, voy a subir los prompts" 1.46M plays; @ray_fu "How to do motion design with Claude Opus 5.5" 147k plays (TikHub).

Everyone was asking "what's the prompt for this video". The author created the repo at 9-27 03:54 UTC, about 5 hours before Musk's repost, right at the peak of demand.

#### 3.2 Biggest launch volume, plus "remake" as backing (the product itself)

- Same-period roundups: athemeroy 9-25 created the repo (per its README, at 10-11 it had 424 source posts, 301 of them Opus); opusvideo/awesome-claude-video 9-25; joeseesun/opus-video-prompts 9-27, with only 18 prompts and 54 cases. This repo launched at 282 entries and hit 475 within two days.

- Other repos just link out or post screenshots. This one embeds a GIF wall right in the README (images hosted on media.skillry.dev), so you see the results as soon as you open it.

- The author's own pitch: "301 Opus 5.5 videos, I spent a full $3,000… and open-sourced all of it" (X post 9-27). The remake edition answers the question readers care about most: "Does this prompt actually reproduce?" No other list has that layer.

#### 3.3 The author's own main account launched it, asking for stars outright (resources + distribution play)

- @yihui_indie posted in Chinese at 9-27 08:16 UTC, 99.8K views, with video. The first reply in the comments was "📦 GitHub open-source repo (would appreciate a Star ⭐️)". TrustMRR lists the author's X following at 64.5k.

- 9-27 The fork accounts that day were mostly Chinese-named users, suggesting the first wave of stars came from the Chinese AI community.

#### 3.4 English AI repost accounts and SEO blogs relayed it (distribution, mostly not paid by the author as far as we can tell)

- First wave, 9-28 to 10-01: frontiernow.dev (9-28), opensourceprojects.dev "475 prompts…" (9-30), artrealmai (9-30), AGI Hunt "Skillry aggregates 389…" (10-01).

- Second wave, early-to-mid October: LinkedIn James Chang (28k followers, around early October), LinkedIn Arthi R. (20k followers, 10-09, "crossed 3,500 stars"), German-language thanisch.co (10-15).

- Peak wave, 10-21 to 10-24:
    - Threads @simplifyinai(10-21)
    - Chen Media long-form article (10-21)
    - LinkedIn Ram Maheshwari "WOW.... this is an insane collection…" (10-22); a Facebook page reposted nearly identical copy ("OMG Dude.... this is an insane collection…")
    - agentcookbook.dev (10-22), agentgid.com "8 Popular GitHub Repos" (10-22), aiidelist (10-24), daily.dev (10-24)
    - An Instagram reel used "Comment 'Viral' for link" to trade comments for the link

This wave lines up with the +1,153 star peak on 10-22.

- Long tail: ailistscout (10-27), Thai-language vibecodingthailand (10-31), charliehills.ai (11-03), Spanish-language codeia.dev, plus trend aggregators like shadowgraph, heatdrop, airanks, repo.tinix.ai, and gitdiscover.

#### 3.5 Why people stayed

It's essentially a copy-paste prompt library: each entry has a source, a category, and technical tags (Canvas / SVG / GSAP / Three.js), plus a three-step "How to use a prompt" guide. Users come to copy a prompt that produces a good clip. From the GitHub side it has no retention problem and doesn't need one. The thing that has to retain users is Skillry's subscription.

### 4. Competitors

| Name | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| lemomo-ai/lemo-opuscar | Open source | 2026-09-26 | 12,987 stars | It's an installable Claude Code plugin with 66 video styles one person tuned, rendered to finished video; this repo is "collecting other people's prompts". The two were the most-starred in the same wave. |
| athemeroy/awesome-claude-5-5-videos | Open source | 2026-09-25 | 963 stars | Launched two days earlier, with research notes and source verification, covering Opus / Sonnet / Fable. No GIF wall, no remake edition, and the author had no big account. |
| joeseesun/opus-video-prompts | open source (Chinese) | 2026-09-27 | 968 stars | Chinese-language, 18 prompts and 54 cases, small scale, no license |
| Vincentwei1021/mg-styles-15 | Open source | 2026-10-02 | 3,190 stars | 15 styles, each with a full prompt and source code, self-produced rather than a roundup |
| zhuyansen/awesome-opus-5.5-video | Open source | 2026-09-27 | 307 stars | A same-day list in the same category, with the same author also maintaining awesome-claude-video-skills (1,668 stars) |
| opusvideo/awesome-claude-video | Open source | 2026-09-25 | 238 stars | A similar list, bilingual Chinese and English. |
| HiAPIAI/awesome-opus-5-5-video-styles | Open source | 2026-09-29 | 54 stars | Re-sorts 425 prompts into 12 art styles, a second-hand reorganization |
| LeaddeOpenLab/awesome-opus-5-5-video-prompts | Open source | 2026-09-28 | 152 stars | Similar lists |
| fable5.app/opus-5-5/works | Web | Not verified | 682 videos | Web-based roundup with a similar catalog size and no GitHub star backing |
| heygen-com/hyperframes | Open-source / big-company tools | 2026-03-10 | 80,024 stars | A framework that renders HTML into video, not a prompt library. Many early entries in the library were produced with it, which the author removed 10-13 |
| yihui-dev/awesome-fable5-5-videos, gemini4-argon, kimi-k3-1, claude-motion | Open source (by the author) | 10-05 to 10-24 | 87 / 33 / 29 / 14 stars | Same template with a different model, none of which took off |

### 5. Differentiation, and whether it's the reason it took off

- **Scale and speed** : the most entries and the most frequent updates in the same time window (batches were added almost daily before 10-13). This is the main reason it became "the Opus video roundup".

- **Original clips side by side with remakes** : the author paid to run 301 to 418 remakes to show the prompts reproduce, which is unique. This helped media and bloggers want to write about it (article titles commonly read "703 Prompts & 418 Remakes"), and more importantly gave the Skillry page content that could catch traffic and convert it.

- **A visual README** : a GIF wall where the results play on open, good for screenshots and sharing.

- **Attribution and links** : every entry links back to the original author's X post. This likely reduced complaint risk and may have made original authors willing to repost (unverified).

Conclusion: differentiation was one reason it took off. The decisive factors were timing, the author's main account launching it, and English repost accounts spreading it organically. The author carried the same differentiation over to Fable 5.5 and other models and only reached double-digit stars, which counts as counter-evidence.

### 6. How many people have seen it

**Revenue is already public and widely discussed.**

- Skillry publicly shows Stripe-verified revenue on TrustMRR: $11,139 over 30 days, MRR $4,647,434 from subscriptions. sellmicrosaas, microsaasexamples, finderlaunch, fazier, and uneed all list Skillry.

- The repo itself was written up by LinkedIn big accounts, Threads, daily.dev, and at least a dozen multilingual blogs.

### 7. Copycats

- Copying the same approach for new models: the author reused this template too; the description of motionvideohq/awesome-fable-5-5-motion-design (10-21) is nearly word for word ("A growing collection of viral motion design videos made with Claude Fable 5.5 and the prompts behind them. Updated regularly").

- Merges or reorganizations: mailes/awesome-claude-5-5-videos ("587 viral videos made with Claude Opus 5.5 & Fable 5.5 — with the prompts behind them"), HiAPIAI/awesome-opus-5-5-video-styles, sunsiyuan/awesome-claude-3d-videos, walidboulanouar/awesome-ai-coded-videos, fable5.app.

- One cycle earlier, when GPT-6 Astra came out, a batch of libraries with the same pattern appeared (magiccreator-ai/awesome-gpt-6-astra with 121 stars, awesome-astra-video-prompts with 18 stars, and an X-RayLuan version that even included "EasyVeo remake prompts"), and none of them took off. This shows people were already running the "new model release → viral prompt roundup" play. What decides the outcome is whether the model produces a visual highlight that floods the internet.

### 8. What others can learn from it

1. **Watch for the visual highlight after a model launch.** Once something like "one prompt yields a stunning result" floods X (this time, Opus 5.5 writing code for motion graphics), whoever publishes the biggest roundup within 48 to 72 hours will see GitHub stars climb on their own.

2. **Pay for "proof of reproducibility"** : the $3,000 remake cost bought credibility that other lists lack, and also gave the site SEO-friendly content pages (one skillry.dev detail page per video).

3. **The repo only drives traffic; the money is collected elsewhere** : every README link carries UTM tags pointing to the author's site, where free prompts and a paid skill subscription sit on the same page, with a Black Friday countdown to push conversions.

4. **Launch from your own big account and ask for stars directly** , then let English repost accounts pick it up organically. These accounts favor libraries with "big number + free + GIF" ("703 prompts… all in one free repo 🤯").

5. **Watch the risks** : 10-27 someone opened an issue pointing out that some prompts contain short links, API keys, and agent configs, and pasting them straight into tool-enabled agents is a security risk. Attribution and copyright with the original creators also need handling (the README says "will delete on creator request").

### 9. Still unclear

- There is no evidence on whether the English repost accounts (Ram Maheshwari, simplifyinai, the Facebook page) shared it organically or were paid.

- The stargazers endpoint 404 can't rule out star-farming. A high fork ratio with very few watchers matches the profile of a bookmarking list, but that doesn't rule out inflated stars.

- How much of Skillry's revenue came from this repo: the timing matches closely (September $1.6k to October $12.2k, with daily peaks lining up), but there is no UTM conversion data.

- Our X corpus (x-radar) only covers up to late September, so reposts on X in October (including the "X trending" mentioned in the data card) can't be quantified.

- Launch dates and traffic for web competitors such as fable5.app have not been verified.

### Sources

- GitHub API: repos/yihui-dev/awesome-opus5-5-videos (stars, forks, commits, issues, forks list), users/yihui-dev/repos, search/repositories

- Data card: /home/ljr/fanout/why/cards/gh_yihui-dev__awesome-opus5-5-videos.json

- Author's X launch post: [x.com/yihui_indie/status/2104122905961595267](https://x.com/yihui_indie/status/2104122905961595267)

- TrustMRR: [trustmrr.com/startup/skillry](https://trustmrr.com/startup/skillry), [trustmrr.com/startup/skillry.md](https://trustmrr.com/startup/skillry.md)

- Skillry: [skillry.dev/ai-videos/opus-5-5](https://skillry.dev/ai-videos/opus-5-5), [skillry.dev/pricing](https://skillry.dev/pricing)

- x library (x-radar): @claudeai 9-22, @shneural 9-24, @ajith_io 9-25, @elonmusk 9-27 (2104135091417215376)

- LinkedIn:Arthi R.(10-09)[www.linkedin.com/posts/arthirajendran_github-yihui-devawesom…](https://www.linkedin.com/posts/arthirajendran_github-yihui-devawesome-opus5-5-videos-activity-7514456294750269440-6le1) ;Ram Maheshwari(10-22)[www.linkedin.com/posts/rammcodes_wow-this-is-an-insane-colle…](https://www.linkedin.com/posts/rammcodes_wow-this-is-an-insane-collection-of-viral-activity-7519062990231453696-q8tE) ;James Chang [www.linkedin.com/posts/james-chang-b6268534_github-yihui-dev…](https://www.linkedin.com/posts/james-chang-b6268534_github-yihui-devawesome-opus5-5-videos-activity-7510323988141641728-sbXS)

- Threads:[www.threads.com/@simplifyinai/post/DewuS2BEWsL/](https://www.threads.com/@simplifyinai/post/DewuS2BEWsL/)

- Blogs: [agentgid.com/guides/opus-5-5-code-videos-8-repos-checked/](https://agentgid.com/guides/opus-5-5-code-videos-8-repos-checked/), [chen.media/guides/703-prompts-behind-viral-videos-that-claud…](https://chen.media/guides/703-prompts-behind-viral-videos-that-claude-opus-5-5-wrote-as-code), [www.opensourceprojects.dev/post/awesome-opus5-5-videos](https://www.opensourceprojects.dev/post/awesome-opus5-5-videos), [agihunt.info/en/p/1a0f7003471d1030cc0c4289919](https://agihunt.info/en/p/1a0f7003471d1030cc0c4289919), [aiidelist.com/blog/awesome-opus-5-5-videos](https://aiidelist.com/blog/awesome-opus-5-5-videos), [charliehills.ai/guides/all-703-prompts-behind-the-viral-clau…](https://charliehills.ai/guides/all-703-prompts-behind-the-viral-claude-videos), [daily.dev/posts/9xlaf4gp3](https://daily.dev/posts/9xlaf4gp3)

- TikTok (TikHub search "opus 5.5 prompts"): @itskeo5, @ray_fu, @benkaluza.lab, and others

- Same-period library report: /home/ljr/fanout/why/out/gh_lemomo-ai__lemo-opuscar.md

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