# Snap a photo, the cup shows up in Blender: a 15-second video made a three-commit repo trend twice · GitHub open source

> Source: https://stringers.io/en/github/gh_ahujasid__camera-to-blender.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 the creator of blender-mcp, which has 33k stars. His "snap a photo and it's in Blender" demo video is instantly understandable and easy to share. September's original post drove the first wave, and in November IG, TikTok, Korean, and Chinese reuploaders reposted the original video for the second.

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

ahujasid/camera-to-blender · ahujasid

# Snap a photo, the cup shows up in Blender: a 15-second video made a three-commit repo trend twice

**What it is** An open source tool: take a photo of a real object with your phone, call Tripo3D to generate a 3D model, and it appears automatically in Blender on your computer within half a minute to a minute.

**Why it took off** The author is behind blender-mcp, which has 33k stars. His "snap and it's in Blender" demo video is instantly clear and easy to share. The September original tweet brought the first wave; in November, IG, TikTok, Korean, and Chinese repost accounts reuploaded the original video and brought the second.

**B** **How you could ride it** Stop building a Blender version. Bring the same "snap and it's in the software" flow to tools like Unity, Godot, Roblox Studio, or Figma. Use open models that run locally to drop the paid API and ngrok. Shoot a 15-second vertical video first, then pitch it to the IG and TikTok accounts that promote GitHub projects.

Grade B: low technical barrier, can be built in a few days, and switching the software and removing the paid API gives it a real edge. The demo video can genuinely be carried by repost accounts. But the first wave ran on the author's own reputation, which followers don't get, so results depend heavily on whether the video gets shared. And once each software vendor builds it in officially, the window closes. Chatter in small circles

Data (updated daily) · Repo created 2026-09-03 · 1,409 stars · 7 days +555

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

The author of blender-mcp, September, posted a demo video: point your phone at something on the desk and take a shot, and half a minute later the same thing appears in Blender. The tweet got 1.41M views, the repo then climbed to about 850 stars, and then sat untouched for two months. In Nov 3, a batch of AI tool repost accounts shared the same video again, and the repo gained over 500 stars in three days. The code hasn't changed a line since Sep 5. This piece covers why it took off, whose capability sits underneath, and which part to study if you want a project that can earn the same stars.

### A two-line tweet

2026 Sep 3, Siddharth Ahuja (@sidahuj on X) posted a tweet with just two lines of text:

> Take a photo of the real world Paste it into Blender 3D

It came with a demo video. The phone points at a physical object, the shutter clicks, and a textured 3D model pops up in the Blender viewport on the computer. By Nov 6, the tweet had 1.41M views, 6,213 likes, 514 reposts, and 5,145 bookmarks. His own X account has about 13k followers, and the views are roughly 100 times his follower count, so the algorithm was pushing it. Bookmarks and likes were about equal, and many people's reaction was "saving this for later."

He's not new. The blender-mcp project he built earlier (now called mcp-for-blender) lets models like Claude operate Blender directly. It has 33k stars and is among the most cited open source projects on "AI + Blender." On Sep 5, AlphaSignal covered the new repo, opening with "the new project from the author of blender-mcp." At that point the repo had only 34 stars.

### What it actually is

Put simply, it's a glue layer. The README lists just three parts:

- A Python relay service running on the computer that calls the various APIs and pushes the model to Blender;

- A phone-based web camera page where you take a photo, confirm, and tap "Send to Blender";

- A Blender add-on that receives the model over WebSocket and imports it into the current scene automatically.

The pipeline: phone takes a photo → Gemini removes the background (optional) → Tripo3D's API generates a textured 3D model from a single image → pushed into Blender, taking 30 to 60 seconds in total.

The barrier isn't low. A Tripo3D API key is required and paid, so every photo burns credits. The phone camera needs HTTPS, so you also have to install ngrok, and on the free plan the address changes on every restart, meaning you reconnect Blender each time. A Korean blogger's long post from Nov 3 put it bluntly: the whole repo has only three commits, and every photo spends Tripo credits. AlphaSignal also noted that the Tripo model version hardcoded in the code now throws errors and needs switching to the v3 endpoint. The repo hasn't been fixed to this day, so cloning it now may not run at all (unverified).

### Two growth waves, with two dead months in between

| Timeline | What happened | star |
|---|---|---|
| 09-03 | Author posts tweet | At 09-05 there were only 34 |
| 09-05 to 09-08 | AlphaSignal, The Daily Commit, VFX DESK, AI accounts on LinkedIn, Weibo, and Chinese AI navigation sites followed up | 09-07 +185, 09-08 +123, reached #12 on Trendshift's JavaScript daily chart |
| Mid-September at 11-02 | No new content, code not updated | 0 to 3 per day, hovering around 850 |
| 11-03 to 11-06 | Repost accounts concentrated on reuploading the original video | 11-04 +230, 11-05 +223, 11-06 +62, reaching 1,368; forks went from 87 to 163 |

(September data from Trendshift; November from daily GitHub snapshots.)

The second wave brought no new features and no new video. It used only the author's September clip:

- On Instagram, @syntaix.ai posted a reel on Nov 3 with the caption "Comment "BLENDER" and I'll send you the GitHub link!" Commenting a keyword and getting the link by DM is a common funnel for AI tool accounts on IG;

- Another IG account, @dealonai.official, also posted it, crediting x.com/sidahuj;

- The Vietnamese TikTok account @cunghocainha posted "Camera to Blender – dự án trending" on Nov 3, with 2,310 views;

- The Korean blog doppingcastle published long-form posts in Korean and English on the same day;

- On Chinese X, @realfxw posted on Nov 4, and @GitHub_Daily (90k followers) relayed it on Nov 5, with 4,172 views.

None of these is large on its own, and no single one explains "+230 in one day." More likely several platforms posted around the same time and their effects stacked (inference; IG view counts couldn't be retrieved). It also wasn't carried over from the author's other project: mcp-for-blender's growth in those days was in line with its usual pace.

### The same thing, already built by others

"Turn a real object into a model in Blender" isn't an idea nobody has tried:

| Who | What it is | Scale | Compared with alternatives |
|---|---|---|---|
| Tripo's official Blender add-on | Image-to-3D and text-to-3D directly inside Blender, available since April 2025 | GitHub 73 star | The backend is identical; it just lacks the "phone photo, automatic push" step |
| Open models like TripoSR, Hunyuan3D-2, and TRELLIS | Locally run image-to-3D models, some with Blender add-ons | 7.3k / 15.3k / 14.2k star | No API fees, but you need a GPU and must import manually |
| Polycam, KIRI Engine, Luma, Apple Object Capture | Photogrammetry scanning apps | Mature product | Shoot dozens of photos around the object; high precision, then export and import into Blender by hand |
| Meshy, Rodin, Tripo web versions | Commercial image-to-3D platforms | Most include a Blender bridge | Same functionality, entered through web upload |
| mcp-for-blender, dream-loop, AnyImage, etc. | Let an agent model in Blender from an image or description | 33.4k / 1.8k / 295 star | Goes through a large model step by step; this one is one shot, instant |

The first line matters most. Same backend: the official add-on has 73 stars after two years, this repo got 1,368. Its model capability is no better than anyone else's, and in practicality it adds two more hassles, ngrok and the relay service. What it adds is one picture: pick up a cup, snap it, and the cup appears in a 3D app. No explanation needed, watched in 15 seconds, and anyone can share it. I also couldn't find a clone with the same functionality on GitHub, and I couldn't check whether any of its 163 forks have changes.

### Why this one, and why then

1. **The author brought his own audience.** blender-mcp gave him a reputation, and the first wave on X reached 1.41M views. Tech coverage all opened with "the author of blender-mcp." Followers can't get that.

2. **Easy to shoot, easy to repost** . Real-world objects turned into software, with a before-and-after anyone gets at a glance, so repost accounts could pick it up two months later and bring a second wave.

3. **Rode the "AI + Blender" wave** . In 2026 September, videos of GPT-6 Astra modeling in Blender were spreading on X, and single-image-to-3D quality had reached the point of "snap a photo and use it as a draft."

4. **Starring costs nothing** . Three commits, MIT license, clear README. Most stars mean "bookmarked, might use later," not that anyone is actually using it.

### What to learn from it

- **Build a demo that explains itself in 15 seconds** . The input should be as everyday as possible (a phone photo, a spoken sentence, a hand gesture), and the output should land directly in a professional tool's interface. The feature can be thin, but the visual has to be understood instantly.

- **A good video can work twice** . The original tweet was the first wave; two months later repost accounts re-uploaded it for another. You can proactively send the video to IG and TikTok accounts that specialize in promoting GitHub projects, in Chinese, Vietnamese, and Korean.

- **Build a name inside an ecosystem first, then ship small tools** . This can't be copied directly, but the principle transfers: if you have no reputation, make your project an add-on to a big project and post it in that project's community.

### If it were you

**Verdict** : worth building a same-category "reality → specific pro software" tool, but don't make another Blender version. The underlying capabilities are all off the shelf; what decides how many stars you get is the demo video and whether people help you spread it.

**Who it's for** : developers who can build small full-stack glue (mobile web pages, local services, plugins for a given app) and cut short videos themselves; ideally someone already active in a software community who knows what its people want.

**Who it's not for** : anyone hoping to make money from this (this kind of project earns stars and reputation); anyone with no distribution channels who won't pursue repost accounts (the author's 1.41M views won't come to you); anyone who wants to build yet another Blender version, since this one and the official plugins already exist.

**How to build it** : swap the output end and connect to tools like Unity, Unreal, Godot, Roblox Studio, Figma, or FreeCAD (whether someone has already done this is unverified); swap the model end to locally runnable ones like Hunyuan3D or TRELLIS, removing the paid API barrier, and make the first line of the README "no API key needed"; drop ngrok in favor of LAN QR-code pairing. Wherever it falls short is your selling point.

**How to get users** : shoot the video before writing the code. A 15-second vertical, single-take clip: pick up an object, take a photo, and it appears in the software. Post on X and the subreddit of the target software; also DM IG and TikTok accounts that promote GitHub projects (including Vietnamese, Korean, and Chinese accounts) and give them the video directly to use.

**When to move** : the sooner the better. The quality of single-image-to-3D is still improving fast, and "snap a photo and it's in the software" still feels new. Once each software's makers build this in natively, the demo won't be impressive anymore (inference: probably a matter of months).

**Signs you've misread it** : a week after posting, the video hasn't passed a few thousand views and no repost account picks it up; or the software you chose already has a one-click photo-to-model feature built in; or the same software already has a similar project with hundreds of stars on GitHub.

[GitHub repo ↗](https://github.com/ahujasid/camera-to-blender)

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

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| Tripo official Blender extension (tripo-3d-for-blender) | Open source | 2025-04 | GitHub 73 star | Same Tripo backend, generating from uploaded images inside Blender; camera-to-blender just adds live phone capture and automatic push, which makes it better for video |
| TripoSR / TripoSG / Hunyuan3D-2 / TRELLIS and their Blender plugins | Open source | 2024–2025 | 7.3k / 1.9k / 15.3k / 14.2k star | Local models: no API fees, but you need a GPU and manual import; camera-to-blender doesn't do modeling, only the photo-to-viewport step |
| Polycam / KIRI Engine / Luma / Apple Object Capture | iOS | 2020–2023 | Mature scanning apps (scale not checked) | Photogrammetry needs dozens of shots around the object for high accuracy, then a manual export into Blender; camera-to-blender takes one photo, fills in the rest with AI, and places it in the scene automatically |
| Meshy / Rodin (Hyper3D) / Tripo web app | Web | 2023–2024 | Commercial image-to-3D platforms, many with Blender bridges | Same function, but the entry point is a web upload; camera-to-blender is open-source glue connecting the phone camera directly |
| mcp-for-blender (by the author himself) and agent-based modeling such as dream-loop and AnyImage | Open source | 2025-03 to 2026-09 | 33.4k / 1.8k / 295 star | Uses chat or reference images to have an agent model inside Blender; camera-to-blender doesn't go through an LLM, one photo is enough |

Where users come from

- **Verified** **author X demo video** : x.com/sidahuj/status/2095512142766342624 (2026-09-03): 1.41M views, 6,215 likes, 5,145 bookmarks; when AlphaSignal 09-05 wrote up the repo, it had just 34 stars, and over the following 09-07 and 09-08 it gained over 300

- **Verified** **Tech media / aggregators** : AlphaSignal, The Daily Commit, VFX DESK, LinkedIn Future AI, Sina Weibo, and Chinese AI navigation sites all covered it between 09-04 and 09-08

- **Verified** **Author already has a reputation (blender-mcp)** : AlphaSignal and doppingcastle both opened with "new work from the blender-mcp author"; mcp-for-blender has 33k stars

- **Verified** **Instagram repost accounts (comment keyword to get the link)** : @syntaix.ai 11-03 reel "Comment BLENDER and I'll send you the GitHub link"; @dealonai.official reel credits via x.com/sidahuj

- **Verified** **TikTok / blogs / Chinese X reposts** : TikTok @cunghocainha 11-03 (2,310 views); Korean blog doppingcastle 11-03; X @realfxw 11-04, @GitHub_Daily 11-05 (4,172 views)

- Inferred **a single trigger for the second wave** : no single big account explains the +230 on 11-04 day; likely multi-platform reposts stacking up

## ahujasid/camera-to-blender: why it works

> Research date 2026-11-06. Channel: GitHub. Data card: last 7 days +514 stars (first 7 days +3), single-day peak +230 (11-04, our database basis), total 1,368 stars / 163 forks, repo created 2026-09-03, last push 2026-09-05.

### Bottom line

blender-mcp (now mcp-for-blender, 33k stars) author Siddharth Ahuja (X @sidahuj) turned "take a phone photo → Tripo3D single-image 3D generation → auto-import into Blender" into a video you understand in 15 seconds 3. His tweet of it got 2026-09-03 1.41M views, and the first wave of stars came from that tweet. From Nov 3 on, AI tool reposting accounts (an Instagram "comment BLENDER for the link" account, Vietnamese TikTok, Korean blogs, Chinese X accounts) mass-reposted the video, bringing a second wave. The code hasn't changed since Sep 5. Both waves of growth came from the video's reach plus the author's reputation in the Blender×AI space.

### What it is

- A very thin glue layer: a FastAPI relay server (`relay_server.py`) + a phone web camera (`webapp/`) + a Blender plugin (receives the model over WebSocket and imports it automatically).

- Flow: phone photo → Gemini background removal (optional) → Tripo3D API generates a preview and a textured model → click "Send to Blender" and the model appears in the current scene; the whole process takes 30–60 seconds.

- Barriers: you need a Tripo3D API key (paid credits, required), ngrok (phone cameras need HTTPS), Python 3.10+, Blender 3.0+. Using a laptop webcam skips ngrok.

- Sources: README (https://github.com/ahujasid/camera-to-blender); AlphaSignal breakdown (https://alphasignal.ai/news/camera-to-blender-turns-one-phone-photo-into-a-3d-asset-in-60-seconds); Korean blog doppingcastle says it has "three commits" and "every photo costs Tripo credits" (https://doppingcastle.com/en/camera-to-blender-photo-to-3d-model/, 2026-11-03).

### Growth curve: two waves, dead water in between

| Period | What happened | Star changes | Source |
|---|---|---|---|
| 09-03 | The author posts the demo video on X: "Take a photo of the real world / Paste it into Blender 3D" | The tweet went out the day it was written, and at 09-05 AlphaSignal's write-up it had only 34 stars; it then shot up | 推文 x.com/sidahuj/status/2095512142766342624:141 万播放、6,215 赞、514 转、5,145 收藏(fxtwitter 接口,2026-11-06 读);AlphaSignal |
| 09-05 ~ 09-08 | AlphaSignal, The Daily Commit, VFX DESK, LinkedIn "Future AI", Chinese AI navigation sites, Sina Weibo and others followed; the author posted on LinkedIn 09-15 too | 09-07 +185, 09-08 +123, Trendshift JavaScript daily chart #12; by mid-September about 840 | Trendshift [trendshift.io/repositories/208277](https://trendshift.io/repositories/208277); Exa search results |
| 09-10 ~ 11-02 | No new content, code not updated | Daily average 0–3, leveling off at about 850 | Trendshift; our GitHub database repo_day |
| 11-03 ~ 11-05 | Repost accounts clustered in bulk (see below) | Trendshift basis 11-03 +264, 11-04 +199, 11-05 +57; our database basis 11-04 +230, 11-05 +223, 11-06 +62; forks 87→163 | Trendshift; data card |

**Who carried the second wave (all secondhand reposts around 11-03, all using the author's original video):**

- Instagram @syntaix.ai posted a reel on 2026-11-03 with the caption "Comment 'BLENDER' and I'll send you the GitHub link!", a classic comment-for-link, DM-to-funnel play (https://www.instagram.com/reel/DfSFPUgJO1c/, comments clustered between 11-03 and 11-06).

- Instagram @dealonai.official posted a reel credited "🎥 via x.com/sidahuj" (https://www.instagram.com/reel/DfRp6fiBbz6/).

- TikTok @cunghocainha (a Vietnamese GitHub-recommendation account) posted "Camera to Blender – dự án trending" on 11-03, with 2,310 views and 77 likes; this account had posted only once before 09-04 (TikHub search).

- Korean blog doppingcastle published long-form posts in Korean and English on 11-03, when the repo had 914 stars.

- Chinese X: @realfxw on 11-04 (968 stars when seen), @GitHub_Daily (91k followers) with 11-05,4,172 views and 39 likes (fxtwitter); the Chinese navigation site ai.lzw.me reposted it 11-04.

- None of these individual posts is large. I did not find a single big account that alone explains the "+230 in one day"; it's more likely that IG reel views plus simultaneous multi-platform reposts together triggered it. Which specific post set off the second wave is **unverified** (IG view counts couldn't be retrieved; the x-radar corpus only covers up to 9).

- It did not come from the author's other projects: mcp-for-blender was steadily adding about +76~118,11-03~05 per day over the same period with no anomaly (our GitHub database repo_day).

### Competitors: who else is doing the same thing

Making a physical object into a Blender model has a lot of competitors:

- **Tripo official Blender extension** (Tripo-Developer/tripo-3d-for-blender, 2025-04, only 73 stars on GitHub): text-to-3D, image-to-3D, and multi-view-to-3D generation directly inside Blender, all on the same backend model. camera-to-blender is basically this with an extra layer on top: a phone camera plus automatic push.

- **TripoSR / TripoSG / Hunyuan3D-2 / TRELLIS** open-source image-to-3D models (7.3k / 1.9k / 15.3k / 14.2k stars, 2024–2025) and their Blender add-ons (e.g. FishWoWater/trellis_blender, 77 stars): run locally with no API fees, but you need a GPU and have to import the results yourself.

- **Photogrammetry / scanning apps** : Polycam, KIRI Engine, Luma, Apple Object Capture / Reality Composer. You shoot dozens of photos around an object to get an accurate model, export GLB/USDZ, then import it by hand (scale not checked, all mature products). Open-source options include SBCV/Blender-Addon-Photogrammetry-Importer (1.4k) and OpenScan.

- **Commercial image-to-3D** : Meshy, Rodin (Hyper3D), and Tripo web apps and apps, all with Blender bridges or plugins; on Gumroad, there's also a paid "Image to 3D Blender Addon" from November.

- **Agent-driven Blender** : the author's own mcp-for-blender (33k), achimala/dream-loop (1.8k, built by 09-07), OmooLab/AnyImage (295), apekshik/anyscene (316), and elithril/blender-kiln (108). All of them go from an image or description to an agent building the scene in Blender.

**How it differs from these** : the model isn't any better than everyone else's, since it's just Tripo underneath. What changes is the input, which becomes a live phone photo, and the output, which lands directly in your open Blender viewport with no exporting or file transfers. The value here is mostly in the **demo** : "pick up a mug, snap a photo, the mug appears in Blender" is instantly clear, which makes it good for short videos. As a practical tool, it actually adds a hurdle compared to the official plugin (you need ngrok and an extra service running).

### Copycats

No matching clone turned up on GitHub (our search covered repos named around photo / camera / phone + blender / 3D, created after 09-03). 163 forks may be doing their own development; unverified. What did show up in the same window are similar ideas, not copies: photo-to-scene agent skills (AnyImage, anyscene, unreal-home-wizard, openhouse-3d), plus Tripo's own growth account TripoGrowthLab (awesome-astra-prompts, 377 stars).

### Why it took off

1. **The author brings his own distribution** : blender-mcp made him one of the best-known names in the "Blender×AI" topic, with 13k followers on X. The original post got 1.41M views, roughly 100× his follower count, which means the algorithm pushed it out for him. AlphaSignal's coverage opened by highlighting "the author of blender-mcp's new project."

2. **Video spreads naturally** : the input is something from the real world, the output is a model made in 3D software, and the before-and-after contrast is strong enough that it needs no explanation. So IG and TikTok repost accounts keep recycling it. The source footage for the second wave is September's original video.

3. **Timing** : 2026 September lines up with the "GPT-6 Astra models in Blender on its own" wave (@Stefan_3D_AI 159k views, @bilawalsidhu, @reach_vb, @npew, per x-radar's September corpus). "AI + Blender" was a hot term on X at the time, and image-to-3D quality had reached the point where a quick snap was usable as a draft.

4. **Low barrier, little code** : 3 commits, MIT license, and a clear README, so the cost of starring it to save for later is low. The high star count (5,145) also suggests many people are saving it "just in case."

### Differentiation, and whether that's why it took off

The differentiation is in the experience layer: **your phone as the camera, plus one-click push into the Blender viewport** . The model layer is Tripo's existing capability with nothing new. This differentiation **really is why it took off, but the mechanism is that it's easy to show** : it packages an existing capability (Tripo's Blender plugin has been around for a while and has only 73 stars) into a 15-second magic video, then spreads it using the author's reputation. Same backend, the official plugin gets almost no views, but this repo reached 1.4k stars. That suggests the stars came from the video and the author's reputation, not from the features.

### What to borrow

- Wrap an existing API as a "reality → software" demo that's instantly understandable. The more everyday the input (phone photos, voice, gestures), the better, and the output should land directly inside a professional tool's interface.

- Build a big project in one ecosystem first to earn a reputation (blender-mcp), then release small tools in the same ecosystem. Each small tool benefits from the big project's reach.

- A good demo video can be used twice: the original post is the first wave, and a reupload by a repost account two months later makes a second wave. You can actively license the video or pitch it to GitHub-recommendation accounts on IG / TikTok, and use the "comment a keyword to get the link" tactic to drive traffic.

- Copying the approach to other platforms also works: swap the output side to Unity, Unreal, Godot, Figma, CAD or Roblox Studio, and the model side to locally runnable ones like Hunyuan3D or TRELLIS, and you drop the paid API barrier.

### Still unclear

- Which specific post sparked the November second wave (IG reel view counts couldn't be retrieved; the x-radar corpus doesn't cover November 3–5).

- Real usage: the tool requires a paid Tripo key, and stars don't mean people are using it. AlphaSignal mentioned that "the hardcoded Tripo model ID now throws 2015 errors and needs an upgrade to the v3 API." The repo still hasn't been fixed, so it may no longer run.

- Whether the author has a partnership or sponsorship with Tripo. Tripo has a dedicated growth team (TripoGrowthLab), and every use brings Tripo API revenue, but I found no evidence of a partnership.

- Whether any of the 163 forks are building improved versions or versions with different models.

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