# "Others built a feature, we open-sourced the whole Higgsfield": a contest tweet with 910k views, and how a local AI film studio gained 376 stars in one week · GitHub open-source project

> Source: https://stringers.io/en/github/gh_drowzeys__Imagination-Studio-by-keys-Wasserman.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

> ComfyUI ran an "open-source Higgsfield alternative" challenge, and Higgsfield sent a lawyer's letter that pushed the contest into the news. Co-author Sam Wasserman posted on the deadline day: "We built the whole Higgsfield as open source, 61+ apps," with about 910k views, and the repo gained 187 stars in a day.

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

drowzeys/Imagination-Studio-by-keys-Wasserman · drowzeys

# "Others built a feature, we open-sourced the entire Higgsfield": a contest tweet with 910k views, and how a local AI film studio gained 376 stars in one week

**What it is** An open-source local AI film studio: script, storyboards, image generation, video with sound, voiceover, music, and editing/export, all in one app. Runs on DGX Spark, 128 GB-class Macs, and RTX Spark.

**Why it blew up** ComfyUI ran a "Higgsfield open-source alternative" challenge, Higgsfield then sent a cease-and-desist that pushed the contest into the news. Co-author Sam Wasserman posted on the deadline day: "We built the whole of Higgsfield as open source, 61+ apps." About 910k views, and the repo gained 187 stars in one day.

**C** **How you could ride it** Don't build another all-in-one AI video studio. Copy its topic and playbook instead: build killer apps for newly launched high-memory local hardware (like Windows on Arm on RTX Spark), or make low-VRAM and commercially usable versions, and post your contest entry while the platform's contest is at peak buzz.

Grade C: the tactics (riding a platform contest, having one number that makes a tweet, attaching a known industry name) are worth learning. But the contest has closed, dozens of entries compete in the same window, the traction came from the event and a filmmaker's follower base that you won't have, and the 128 GB requirement caps the ceiling. What people in small circles are saying

Data (updated daily) · Repo created 2026-10-04 · 617 stars · 7 days +165

GitHub · stars

Data 2026-10-09 → 2026-11-08 (31 days, 30 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Imagination Studio is an open-source "AI film studio" built by two people. Scripting, storyboarding, image generation, video with audio, voice-over, music, editing and export all run in one local app that requires a 128 GB-class high-memory machine. It picked up over 200 stars on GitHub in a month, then rode ComfyUI's #OpenSourceHiggsfield challenge and a cease-and-desist storm. On the deadline day a co-author posted a contest entry, and it gained 376 stars in a week. The product is genuinely complete, but the thing that pushed it has already passed, and the people who followed along can't be borrowed.

### The deadline-day post

On Oct 2026 30, Emmy-winning filmmaker Sam Wasserman tagged ComfyUI on X with #OpenSourceHiggsfield and posted this:

> "We didn't just build 1 features, we built the whole Higgsfield platform and more, and it's open source. 🎬 Imagination Studio. An all-in-one studio with 61+ apps: video, music, audio, sound effects / foley, agent, prompt assistant, depth maps, stem separation, LoRA, custom 3D previs, editing."

By early November, the post had about 913k views, 1.5K likes, 162 reposts and 2.4K bookmarks (archived at whatpeoplemade.com). More bookmarks than likes means most people saved it to install later.

The repo went from 216 stars in Oct 30 to 265 stars by Oct 31, then gained 187 stars in a single day on Nov 1, reaching 452. Forks also jumped from 24 to 72. As of Nov 6,592: 92 stars and forks (GitHub, 11-06).

### What it actually is

The README's first line: "A local film studio: plan a film, generate visuals, video, voice and music, then edit and export."

- **Script and production:** import scripts, break down shots, maintain a "film bible", build storyboards. This part is called Slate, designed by Sam Wasserman after how real film crews work (casting, storyboarding, first AD).

- **Visuals and video:** images with Qwen-Image-2.1 and FLUX3; video with MiniMax H3, open weights released in late July, output comes with sound.

- **Sound:** multi-speaker dialogue voice-over, songs, sound effects and foley.

- **Finishing:** organize footage, color grade, mix audio, make animated subtitles, export to editing software.

Everything runs on your own machine, no cloud API. Underneath is TensorFold, an inference engine the authors wrote themselves. The requirements are spelled out plainly: 48 GB for image-only on Mac, 96 GB for the standard video install, 128 GB for full features, plus 185 GB of disk. The Linux side targets NVIDIA DGX Spark; in early November they added a native one-click install for Windows on Arm with RTX Spark N1X.

The two authors each cover half. keys (GitHub drowzeys, X @u1tra_instinct) is a familiar name in DGX Spark local inference circles, known for deployment guides. Sam Wasserman has an IMDb credit and has appeared on AI film podcasts. One writes the code, the other brings the process and the name.

### Where the stars came from

| Timeline | What happened | Stars (GitHub) |
|---|---|---|
| 10-04 | Repo created | — |
| 10-15 | ComfyUI launched #OpenSourceHiggsfield with a $25k prize | — |
| 10-16 ~ 10-18 | One user shot a 2-minute short with it and opened about 10 issues in one go | 103 → 137 |
| 10-21 | Sam posted "Imagination Studio is Out Now"; several AI film sites picked it up | 143 → 156 |
| 10-23 ~ 10-28 | 3.2.0 "a film studio from start to finish"; a Chinese account, AGI Hunt, wrote about making a short on a single DGX Spark | 182 → 213 |
| 10-29 | Higgsfield sent ComfyUI a cease-and-desist, ComfyUI publicly refused, and the contest made the news | 216 |
| 10-30 | Released 4.3.0 "Apps", packaging features into 61 one-click apps; Sam posted his contest entry the same day | 216 |
| 10-31 ~ 11-02 | Russian, English and Chinese AI news sites and newsletters like AlphaSignal wrote up "61 apps" | 265 → 452 → 499 |
| 11-03 | Contest deadline (extended to this day); an AI film site listed it among five contest entries worth watching | 517 |
| 11-06 | Still gaining 17~30 stars a day | 592 |

In its first month it rode word of mouth in local-inference and AI-film circles, gaining a few stars a day. The later surge was almost entirely one event: the contest gave it a ready-made tag, the cease-and-desist pushed attention to its peak a day before the deadline, and Sam's post landed right at that moment.

Two details were clearly planned by the authors. One: the "4.3.0 apps" release came the same day as the post, repackaging existing features into 61 one-click apps, so the "61+" number held up. Two: right after it blew up they quickly followed with a native Windows version, catching people who had just bought RTX Spark.

Channels that didn't work: there was no Hacker News post, no dedicated Reddit thread turned up in search, and no mainstream tech press coverage.

Fork rate is about 15.5%, high for this scale (inference: many of the arrivals are hardware enthusiasts who actually own DGX Spark or high-memory Macs and will install it).

### Who else is doing the same thing

That one contest alone produced dozens of "open-source Higgsfield" projects.

| Project | Scale | Compared with alternatives |
|---|---|---|
| Higgsfield | Commercial AI video platforms | The thing being replaced. Imagination Studio is free and not pay-per-use, but needs 96~128 GB of memory |
| D3voz/openjutsu | 1,729 stars, +7 in 1,433 days | An entry from ComfyUI's official team that only does character swapping, and gained far fewer stars than this one |
| wide-trace/open-higgsfield | 5,342 stars | Available since August, one prompt box for multiple models, generation-focused |
| wexloops/openfield | 172 stars | Runs on ComfyUI, works with a 24 GB GPU, strong on camera and motion control |
| FlashML-org/FreeVideo | 2,778 stars | The opposite direction: running MiniMax H3 in 8 GB of VRAM |
| Anil-matcha/Open-Generative-AI | 32K stars | A multi-model generation workbench for 600 models, mostly through APIs |
| Toonflow, waoowaoo, ArcReel and others | 2k~20k stars | Chinese short-drama and comic-drama pipelines, mostly connected to cloud models |
| IV Studio and other entries from the same contest | Few stars | Also "idea → script → storyboard → MiniMax H3 final cut", the closest match |

What sets it apart is a full filmmaking pipeline with its own engine, entirely local. Most entries in the same contest handle one stage or depend on ComfyUI. But this difference affects whether it gets written up and whether reviews hold up, not why it grew: in the same contest, openjutsu, which only does one feature, grew faster thanks to ComfyUI's official account.

### Its weak spots

- **The barrier is the ceiling.** Few people worldwide have 128 GB-class machines; many stars are probably bookmarks, and far fewer people have actually installed it than 592 (inference).

- **The model licenses block commercial use.** MiniMax H3's community license excludes the US, EU, UK and South Korea, and Qwen-Image-2.1 weights are non-commercial only. That stops it from being a real production tool.

- **The buzz hangs on the contest.** Judging happens in November 10, and a win could bring another wave; without one, it will probably drop back to ten or twenty stars a day.

### If it were you

**One-line verdict:** you can learn from it, but don't copy the topic. The "open-source Higgsfield" wave has closed, dozens of entries compete for the same attention, and its growth came from a platform event and an Emmy-winning filmmaker's following. You won't have either.

**Who it suits:** developers who already own high-memory machines like DGX Spark, RTX Spark or M5 Ultra, are already in local inference circles, and can get an open model running smoothly.

**Who it doesn't suit:** anyone planning another "all-in-one AI video studio"; several projects on that path already have thousands of stars. Also anyone without the matching hardware to test things themselves.

**How to do it:**

1. Take its topic-picking logic, not its topic: build the killer app for newly launched, high-end local hardware that people "bought but don't know what to do with." RTX Spark N1X only went on sale in October, and there are still few AI apps that run natively on Windows on Arm with an NVIDIA GPU. That's a bigger gap than another film studio.

2. Or fill the end it skipped: it only serves 128 GB machines, while FreeVideo shows that "squeezing into low VRAM" alone can earn two or three thousand stars. Likewise, a compliant version built only on commercially usable open models would fix its license weakness.

3. Time the launch to the spread: prepare one concrete number that fits in a tweet (its "61 apps"), and ship on the same day you post.

**How to get users:** watch the platform challenges, hackathons and hot controversies that ComfyUI, model makers and hardware vendors run. During the peak days, post your contest entry with a line like "others built one, we built them all" and tag the organizers. Find a known name in the field to co-sign: they provide the story, you provide the code. Then pitch to NVIDIA developer forums and relevant newsletters. HN is worth a post, but don't count on it.

**When to move:** this round is over. The next chance will follow the next contest or the next hardware launch; the high-memory local machine wave probably has a few months left (inference).

**Signs you misread it:** after the Nov 10 awards, it got a second wave, and a batch of "local film studio" copycats also picked up stars, which suggests the topic itself is bigger than the contest buzz. Your project has been out for two weeks, the organizers and the relevant accounts haven't reposted it, and the stars are stuck in double digits. That means you didn't ride the event, and a complete product alone isn't enough.

[GitHub repo ↗](https://github.com/drowzeys/Imagination-Studio-by-keys-Wasserman)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| Higgsfield | Web | 2024 | Commercial AI video platforms | It's the thing being replaced. IS is free and local, not pay-per-use, but needs 96~128 GB of memory |
| D3voz/openjutsu (ComfyUI team) | Open source | 2026-10-26 | 1729 stars, +7 in 1433 days | Only does character swapping, relies on ComfyUI; IS is a full filmmaking pipeline |
| wexloops/openfield | Open source | 2026-10-28 | 172 stars | Runs on ComfyUI, works with a 24 GB GPU, strong on scheduling and shots; IS has its own engine and targets high-memory machines |
| wide-trace/open-higgsfield | Open source | 2026-08-26 | 5342 stars | Prompt-box multi-model generator; IS leans director / producer workflow |
| Anil-matcha/Open-Generative-AI | Open source | 2023-05 | 32K stars | 600 multi-model generation workbench, mostly via API; IS runs fully local |
| SegFault42/HeliosGen | Open source | 2026-04 | 3295 stars | Self-hosted desktop generator, no film workflow |
| Toonflow / waoowaoo / ArcReel / LocalMiniDrama | Open source | 2026-01~02 | 2k~20k stars | Built for Chinese short dramas and comic dramas, wired to cloud models; IS is for English filmmakers, running entirely on local hardware |
| FlashML-org/FreeVideo | Open source | 2026-09-13 | 2778 stars | Squeezes MiniMax H3 into 8GB VRAM, video generation only; IS loads up big RAM and goes full-featured |
| ComfyUI_MiniMaxH3_Director / Rapid-H3 | Open source | 2026-08~10 | 2699 stars / not indexed | ComfyUI nodes, not a finished app |
| IV Studio / Kitty Open Studio / Comfy Previs / Cheapsfield | Open source | 2026-10 | Same-contest entry, not indexed | Most cover just one stage, or depend on ComfyUI or Comfy Cloud; IS is an end-to-end pipeline with its own engine |

Where users come from

- **Verified** **X contest post (Sam Wasserman)** : 10-30 @ComfyUI #OpenSourceHiggsfield post, about 913K views, 1.5K likes, 2.4K bookmarks; repo gained 49 and 187 stars from the next day

- **Verified** **Platform event: ComfyUI and BFL's $25k challenge, plus the Higgsfield cease-and-desist** : blog.comfy.org launched it on 10-15; 10-29 cease-and-desist, Comfy refused; thereelstack 11-03 named it one of five entries worth watching

- **Verified** **Secondary spread via AI news sites and newsletters** : hermes-ai.net (utm_source=alphasignal), cpu3d in three languages, uhnivenko, AGI Hunt, whatpeoplemade, The Shiplog, ArtRealmAI, AI Film History

- **Verified** **Author network in the DGX Spark / local inference scene** : keys has 70 multiple Spark recipe repos, credited by other people's repos; 14 external contributors; fork rate 15.5%

- **Verified** **Industry reputation of the co-author** : Sam Wasserman is an Emmy-winning filmmaker with an IMDb entry and podcast appearances

- Inferred **HN / Reddit** : 0 entries in the hackernews library; no dedicated Reddit thread found

## Imagination Studio by keys + Wasserman: why it took off

Repo: [github.com/drowzeys/Imagination-Studio-by-keys-Wasserman](https://github.com/drowzeys/Imagination-Studio-by-keys-Wasserman) (Apache-2.0, Python, created 2026-10-04). Data (github repo_profile / data card, 2026-11-06): 592 stars, 92 forks (fork/star about 15.5%). Last 7 days +376, first 7 days +60; biggest single day +187 (11-01). GitHub API reading that afternoon: 601 stars, 93 forks.

### One line

This is an "AI film studio" that runs on your own machine: one app covering script, storyboard, image generation, video with sound, voiceover, music, editing and export, built on the TensorFold inference engine. Target hardware: NVIDIA DGX Spark, 128 GB-class Apple Silicon Macs, and later Windows on Arm with RTX Spark N1X. Two makers: keys (drowzeys / X @u1tra_instinct), who builds deployment recipes in the local DGX Spark inference scene; and Sam Wasserman, an Emmy-winning filmmaker who made the "Slate" production workflow.

### Verdict

It mostly took off on one thing: ComfyUI's **#OpenSourceHiggsfield** challenge. 2026-10-15 launched with a $25k prize; 10-29 Higgsfield sent a cease-and-desist, ComfyUI publicly refused, and the contest became news. On the 10-30 deadline day, Sam Wasserman posted his entry: "We didn't build 1 feature. We open-sourced the whole Higgsfield platform, 61+ apps." The post drew about 913k views, 1.5K likes and 2.4K bookmarks (archived on whatpeoplemade.com). The repo then gained 49 and 187 stars on the next two days.

The product also has substance: it isn't a single feature but the full pipeline from script to finished film, all running locally. That fits both the "open-source alternative to Higgsfield" story and the crowd running MiniMax H3 on their own cards.

### Timeline and growth evidence

| Date | Event | Stars (github repo_day) |
|---|---|---|
| 10-04 | Repo created (keys already had 70 DGX Spark deployment recipe repos, e.g. the MiniMax-H3 single-Spark acceleration repo with 49 stars) | — |
| 10-04 | Listed on artificialbuilds.com | — |
| 10-15 | ComfyUI contest #OpenSourceHiggsfield launched (blog.comfy.org) | — |
| 10-16 | Earliest visible data | 103 |
| 10-18 | BTankut made a 2-minute short and opened about 10 issues in one go | 137 |
| 10-20 | keys posted praising Sam's studio (nitter archive) | 142 |
| 10-21 | Sam posted "Imagination Studio is Out Now"; ArtRealmAI and AI Film History covered it | 143 → 156(10-22) |
| 10-23~28 | 3.2.0 "film studio, start to finish"; AGI Hunt wrote up a short, 'Daybreak', made on a single DGX Spark | 182 → 213 |
| 10-29 | Higgsfield sent Comfy a cease-and-desist, Comfy publicly refused (thereelstack, techcityauthority) | 216 |
| 10-30 | 4.3.0 "Apps": 61 one-click apps; same day Sam posted the @ComfyUI #OpenSourceHiggsfield entry (about 913k views) | 216 |
| 10-31 | uhnivenko, cpu3d (Russian / English / Chinese), and AGI Hunt followed with "61 apps"; fork timestamps got denser from 10-30 20 (UTC) | 265(+49) |
| 11-01 | Peak | 452(+187) |
| 11-02 | hermes-ai.net (indexed from AlphaSignal), The Shiplog coverage | 499 |
| 11-03 | Contest deadline (extended to this day); thereelstack named it one of five entries worth watching, calling it the "MiniMax H3 film studio" | 517 |
| 11-04 | keys posted: RTX Spark N1X native Windows one-click app (3,283 views) | 543 |
| 11-06 | Still gaining 17~30 stars a day | 592 |

### Where users come from (acquisition channels)

1. **Contest post on X** (confirmed): Sam Wasserman's #OpenSourceHiggsfield entry, 10-30, about 913K views, 1.5K likes, 162 reposts and 2.4K bookmarks. Source: [www.whatpeoplemade.com/i/imagination-studio-an-open-source-a…](https://www.whatpeoplemade.com/i/imagination-studio-an-open-source-all-in-one-ai-creative-778021.html); original post: [x.com/SamJWasserman/status/2116095035720778021](https://x.com/SamJWasserman/status/2116095035720778021). The timing lines up with the 11-01 peak.

2. **Platform event** (confirmed): ComfyUI and BFL's $25k challenge and the Higgsfield cease-and-desist. Media covered the entries closely (thereelstack.com has several pieces), and the "open-source Higgsfield" angle carries its own traffic. The sister project D3voz/openjutsu, built by the official ComfyUI team, got 10-26 stars in 7 days, 1433 stars, showing the buzz was shared across the whole contest track.

3. **Secondary spread via AI news aggregators and newsletters** (confirmed): AlphaSignal (hermes-ai.net links carry utm_source=alphasignal), cpu3d.com in three languages, uhnivenko.com, AGI Hunt (Chinese and English), whatpeoplemade, The Shiplog, ArtRealmAI, AI Film History.

4. **Old ties in the DGX Spark / local-inference scene** (confirmed, smaller effect): keys is a familiar face in the DGX Spark scene. Other people's repos credit him in CREDITS (e.g. tonyd2wild), and there are NVIDIA developer forum threads on TensorFold. Contributors include same-scene names like MiaAI-Lab, davidlowjw and ivberg (who worked on the Windows ARM64 port). A fork rate of 15.5% is high for this size, suggesting many visitors are hardware tinkerers who actually install it.

5. **The creator's own reputation** (confirmed): Sam Wasserman is an Emmy-winning filmmaker with an IMDb entry, and has appeared on podcasts (We the Film3, a Matt Workman interview). "A workflow designed by a real filmmaker" is a selling point in itself.

6. Not found: HN (0 entries in the hackernews library), dedicated Reddit threads, paid promotion. The x-radar local corpus also doesn't index its related tweets.

### Competitors

| Name | Where | Launched | Scale | How it differs |
|---|---|---|---|---|
| Higgsfield | Web (commercial SaaS) | 2024~ | Big-tech-grade AI video platforms | It's positioned as an "open-source alternative": Imagination Studio is free to run locally with no per-use fees, but the hardware bar is 96~128 GB |
| D3voz/openjutsu (ComfyUI team) | Open source | 2026-10-26 | 1729 stars, +7 in 1433 days | openjutsu only does character swapping, and still depends on ComfyUI; IS covers the whole pipeline from script to finished film |
| wexloops/openfield | Open source | 2026-10-28 | 172 stars | Runs on top of ComfyUI, usable on a 24 GB GPU, strong on orchestration, shots and motion; IS ships its own engine and targets large-memory machines, with broader coverage |
| wide-trace/open-higgsfield | Open source | 2026-08-26 | 5342 stars | One prompt box connected to multiple models, generation-focused; IS leans toward director and producer workflow |
| Anil-matcha/Open-Generative-AI | Open source | 2023 | 32K stars | 600 multi-model generation workbench, mostly via API; IS runs fully local |
| SegFault42/HeliosGen | Open source | 2026-04 | 3295 stars | Self-hosted desktop generator with no Slate-style filmmaking workflow |
| HBAI-Ltd/Toonflow-app, waoowaoo, ArcReel, LocalMiniDrama | Open-source (Chinese short dramas) | 2026-01~02 | 2k~20k stars | Built for Chinese short dramas and comic dramas, mostly wired to cloud models; IS is for English filmmakers on local hardware |
| storytold/artcraft | Open source | 2022 | 39K stars | Lean toward 3D and canvas-based creation engines |
| FlashML-org/FreeVideo | Open source | 2026-09-13 | 2778 stars | Gets MiniMax H3 running on 8 GB of VRAM for single-purpose video generation; IS goes the other way: loads up big RAM and goes full-featured |
| AIMixer/ComfyUI_MiniMaxH3_Director, Rapid-H3 | Open source (ComfyUI node) | 2026-08~10 | 2699 stars / not indexed | Node-level tools, not finished apps |
| IV Studio, Kitty Open Studio, Comfy Previs, Cheapsfield, OpenIllusion | Open-source (same contest) | 2026-10 | Not indexed, or very few stars | Same-category entries from the same contest; IV Studio also runs "idea → script → storyboard → MiniMax H3 film", the closest match to IS |
| Unsloth Studio | Open source | Old Library | 79K stars | Local UI runs MiniMax-H3, but the focus is LLM training |

Verdict: lots of competitors. Just in this one contest there are dozens of "open-source Higgsfields." It shipped not because there was no competition.

### Why it took off

- **Rode a platform event** : ComfyUI's challenge offered ready-made tags, prize money and media attention, and the lawyer-letter drama pushed buzz to its peak one day before the deadline. Sam's entry post went up on the 10-30 deadline day, perfectly timed.

- **Narrative maxed out** : "Others built 1 features, we built the whole Higgsfield, 61+ apps." The numbers sell. Bookmarks (2.4K) outnumber likes (1.5K), the classic "save it for later." 4.3.0's Apps directory shipped the same day, clearly meant to make the "61 apps" claim hold up.

- **Complete product** : script, casting, storyboard, images, video with sound, voiceover (16-person dialogue), music, Foley, editing, color grading, export to NLE, all in one local app. Most peers cover just one stage.

- **Complementary authors** : keys brings hardware and inference-engine chops (TensorFold, DGX Spark deployment recipes), Sam brings film-industry workflow and name recognition. Both circles bring people in.

- **Timing** : MiniMax H3 (2026-07-30 open weights, video with sound), Qwen-Image-2.1 and FLUX3 open models just matured; machines with 128 GB unified memory like the DGX Spark and M5 Ultra, plus the RTX Spark N1X launching in October, made "shooting films on one machine" feasible for the first time. People who bought this hardware need a killer app.

- **Fast iteration** : went from 3.1 to 4.3.1,212 PRs in one month, rapidly followed the burst with a native Windows version, and picked up 14 external contributors.

### Differentiation

- Fully local, with its own engine (TensorFold: written in Zig, on Metal / CUDA). No dependency on ComfyUI node graphs, no cloud API.

- It's a "director / film crew workflow" (Slate: film bible, casting, storyboards, first assistant director), not "prompt box + model dropdown."

- Unlocks features based on machine memory, with real-world eligibility checks, and only accepts large-memory machines. That's both a moat and a ceiling (Macs need 96 GB minimum).

These differentiators are why it can "stick around" and get written about, but they didn't drive the growth. The growth came from the contest and that tweet. In the same contest, openjutsu grew more on differentiation alone (thanks to ComfyUI's official accounts).

### Visibility

No revenue (open source). Seen by about 910k people on X, covered by 8+ AI news sites and newsletters, and listed in the jutsuhq/open-source-higgsfield roundup. It spread widely in the "AI video / local AI" circles, but it hasn't hit HN and no mainstream tech outlet has written about it.

### What others can borrow

- Target the platform's challenges or hot controversies, and post an entry tweet on deadline day with a punchy line like "we built all of it."

- Release timing should follow the spread: 4.3.0 repackaged existing features into 61 one-click apps, giving the tweet a concrete number.

- Pair a technical author with an industry name: one ships the code, the other supplies the narrative and trust.

- Target newly launched high-end local hardware (DGX Spark, RTX Spark, M5 Ultra) and build the app for owners who don't know what to do with it yet.

### Still unclear

- 11-10 judges the contest; a win could bring another wave. Without a win, growth may settle back to 10~20 stars a day.

- Few people have 128 GB-class machines, so real installs may be far below the star count. How many stars are "bookmarked and never installed" is unverified.

- GitHub REST rate limits, and the GraphQL stargazer list came back empty, so I couldn't check star timestamps hour by hour. Peak attribution relies on fork timestamps lining up with the tweet time.

- The MiniMax H3 community license excludes the US, EU, UK and South Korea, and Qwen-Image-2.1 weights are non-commercial only. Commercial use is blocked, which hinders its path to being a real production tool.

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