Over the years, I have followed various clock makers online, looking at their designs, 3D printed models, and generally spending more time than I would like to admit trying to get a 3D printed clock running. Makers such as the excellent Brian Law (Wooden and 3D printed), Steve’s Clocks (great insights and designs), JBV (amazing, complex engineering) and Wooden Gear Clocks (a lovely site for premade clocks which come in kits but still need more skills then it turns out i have in basic cutting/sanding of brass rods) have all inspired me, but have also driven my need for a simpler, open source design which would be free and easier to build.

It has taken a year to perfect, mainly as a summer project last year (in between work) and refining it this year to reach a point where other people could make it.

All 3D printed clocks are however never completely ‘easy’ to build, they all need a few extra parts, mainly in the need to reduce friction with metal rods, bearings, but we have limited the parts to the minimum, made sure they are all off shelf parts and have also reduced the need for any sanding/cutting/ beyond one small part. We have also provided all the files available on Printables, and incoming to Github, allowing others to refine and contribute new designs or updates. As time goes along, the clock will evolve – but for now it’s ready for its first release, and it’s called TIME.

We have tried and mainly failed with other designs online; as such, we wanted to build our own version, from first principles, not modifying others, but from the ground up, with the following aims:

1. It should be as easy to build and replicate as possible

2. Any additional parts should be easy to source and low-cost

3. Cutting/Sanding should be limited

4. It should have a proper 1-second tick/tock sound – this was important.

As such, the first thing to learn was how an Escapement Mechanism worked, how it contributes to the timing of a clock, the importance of the length of the pendulum and how it dictates the sound of the clock.

First Steps: The Escapement

While most people listen to music on their headphones on the way to work, last summer, I started listening to ChatGPT’s newly introduced voice mode to talk me through the history and detailed workings of a clock escapement mechanism. It allowed me to gain enough knowledge to start drawing my own deadbeat escapement.

A clock’s escapement is the heart of its mechanism, ingeniously translating the constant power from the gear train into the precise, rhythmic pulses that create the “tick-tock.” It performs two critical jobs: it allows the gears to “escape” forward one tooth at a time, regulating the speed of the hands, and it gives the pendulum a tiny push on each swing to overcome friction and keep it moving. The deadbeat escapement, perfected by George Graham around 1715, was a major leap in accuracy. Unlike earlier “recoil” escapements, where the escape wheel would kick backwards slightly after each tick, the deadbeat’s pallets are shaped so the teeth land “dead” with no recoil. This crucial improvement prevents the escapement from disturbing the pendulum’s natural, isochronous swing, making the clock significantly more accurate and establishing the design as the standard for precision regulator clocks. My 3D tool of choice to design mechanisms/enclosures for devices is Autodesk 360, as such, all I needed was a reference drawing and the thought that getting a working escapement would be a good first step. Over at Abbey Clock there is an excellent guide on Drawing Graham Pallets – which led to our own slightly modified design and, as pictured below, our first working escapement:

We haven’t published our prototypes yet, but I can share them if people are interested in either the escapement above or the next stage – the Pomodoro Timer, below. The escapement mechanisim is powered by a weight and counterweight sytem with a 1 metre pendulum – this provided a way to perfect the initial ‘tick tock’ of the clock, designed to provide a steady beat every 0.5 seconds, allowing the escapment to move forward, once every seconds and more importantly, make a full rotation once a minute.

Gear Ratios: The Pomodoro Timer

The second step was to add an additional gear and gain an understanding of gear ratios. Gear ratios for a clock are all-important as they not only define the number of gears you need, but also define the relationship between them, turning the fast-paced energy of the escapement into the slow, readable passage of time.

At its core, a gear ratio translates speed and torque between rotating shafts. In our clock, we have a known starting speed: our escape wheel’s shaft rotates once every minute (60 seconds). To build a timer that rings a bell every 25 minutes, we needed to create a gear train that would complete one full rotation in that time.

The maths to figure out the required gear ratio is straightforward:

• Target Rotation Time: 25 minutes = 25×60=1500 seconds.

• Source Rotation Time (Escape Wheel Shaft): 60 seconds.

• Required Gear Ratio: Target Time​=601500​ or 25:1 – as we see later on, it is the ratio which is important, and arguably, easier to understand.

As you can see in our prototype, we achieved this with a compound gear train made of two identical stages, which makes the design elegant and easy to replicate. A compound gear it a cluster of two or more gears of different sizes that are fixed together on the same shaft, forcing them to rotate at the same speed. In short, instead of having one massive gear drive a tiny one to get a big gear ratio, a compound gear lets you achieve the same result in stages.

This setup is the key to creating a gear train that can achieve a large change in speed or in a compact space

1. First Stage: The shaft from the escapement has an 8-tooth pinion that drives a 40-tooth gear. This gives a reduction of 840​=5:1.

2. Second Stage: Mounted on the same shaft as the first 40-tooth gear, a second 8-tooth pinion drives the final 40-tooth gear. This provides another reduction of 840​=5:1.

The total reduction is the product of the individual stages: 5×5=25:1. This ratio perfectly transforms the 60-second rotation of the escapement shaft into the 25-minute rotation needed to trigger the bell. Or at least that’s how it should be in a perfect world. To be honest, I experimented a little, and my timings were a little out, coming in at approximately 20 minutes per bell ring, with the weight running the timer for an hour. The main point is that I had extended out from the escapement and used compound gears to start using ratios for timings. Of note, Chat GPT was useful in the wider understanding of the clock mechanism and theory, but it would frequently miscalculate gear trains and ratios, so the final design was done with old-fashioned logic.

TIME: 3D Printed Clock

The final clock is simply a case of building out the number of gears, using the same logic as the Pomodoro Timer. I had an escapement rotating once a minute, and I needed the main gear rotating once an hour, so I could attach an hour hand to it – that’s a ratio of 60:1

Following that logic and a ratio of 60:1 –

• Escapement: The Minute Gear – The escape wheel has 30 teeth, with a 10-tooth pinion on its shaft, delivering a 1-second “tick-tock” rhythm. The escape wheel completes one full rotation every minute.

• Gear 1: 40-tooth wheel (driven by the 10-tooth pinion) and 20-tooth pinion.

• Gear 2: 60-tooth wheel (driven by the 20-tooth pinion) and 24-tooth pinion.

• Gear 3: The Hour Gear- a 120-tooth wheel (driven by the 24-tooth pinion), which completes one rotation per hour and connects to a drive gear (details on the drive gear to follow).

The gear ratios provide the necessary speed reduction to convert the escapement’s motion into hourly rotation. Starting from the escapement:

• The 10-tooth pinion drives the 40-tooth wheel of Gear 1, creating a 40:10 (or 4:1) reduction ratio—Gear 1 rotates at 1/4 the speed of the escape wheel.

• Gear 1’s 20-tooth pinion drives Gear 2’s 60-tooth wheel, a 60:20 (or 3:1) ratio—Gear 2 rotates at 1/3 the speed of Gear 1.

• Gear 2’s 24-tooth pinion drives Gear 3’s 120-tooth wheel, a 120:24 (or 5:1) ratio—Gear 3 rotates at 1/5 the speed of Gear 2.

It all suddenly seems complicated – but if you look at it more simply, the cumulative reduction ratio is 4 × 3 × 5 = 60:1. Since the escape wheel rotates once per minute, Gear 3 rotates once every 60 minutes—or exactly once per hour.

I could have chosen any ratio for the gears as long as it works out to 60:1.

With a main gear turning once an hour, upon which an hour hand can be attached, all that is now needed is a Drive Gear to power the clock and a small subset gear on which to put the minute hand.

The Drive

The drive gear serves a dual purpose as a catch mechanism, enabling the clock to be wound up while preventing unintended unwinding. This gear, connected to the drum, includes a ratchet and pawl system. When winding the clock, the ratchet allows the drum to rotate in the winding direction, lifting a weight of 2kg. The pawl engages the ratchet teeth, locking the drum in place to stop it from unwinding backwards once the winding is complete. This ensures the stored energy remains secure, releasing only through the controlled escapement and gear train during operation.

The Minute Hand

Finally, there is a separate gear chain for the minute hand, which fits behind the hour hand; this allows the traditional hour and minute hand configuration (the hour hand fits on a brass rod going through the hour gear, holding everything in place while also allowing it to freely move.

And that’s it! It seems simple when you break it down, although designing and building it from scratch took a little more time than I thought, and I lost count of the iterations it took to get here. The design will continue to be refined—perhaps with the addition of an hourly chime in the near future.

For now, we hope you enjoy making the clock. Do let us know if you build one, over at Printables (the 3D printed files are incoming, we are just collating the files)…

The post TIME: An Open Source 3D Printed Clock appeared first on Digital Urban.

Everything Changed

The term ‘Vibe Coding’ was defined Andrej Karpathy, a co-founder at OpenAI in Early 2025. The basic idea is to rely entirely on Large Language Models – LLMs – and code by using only natural language prompting, rather than writing the code yourself.

Back in 2023, I published ‘Frame-IT’, an iOS app on the Apple Store, written in Swift but coded completely in the then-early version of ChatGPT. As I noted on DigitalUrban, every app has a story, from the idea through to story boarding, finding a design team, working with in house computer scientists or outsourcing. All of these steps take time, resources and often for the small ‘would be developer’ the obstacles are overwhelming, meaning that the spark of an idea often gets lost.

Yet, everything has changed, the whole landscape of developing, designing and creating has turned on its head. It has reached the point where the spark of an idea can be coded, designed, marketed and launched with the help of Artificial Intelligence. Frame-IT is a very simple app, doing a simple thing, but it came from an idea, a want, a desire to do something which before AI would have got lost in logistics, finding the time of computer scientists, and it would certainly be economically unfeasible.

The First Idea

Frame-IT came about while looking at Samsungs’ Frame Televisions, beyond the normal 4K television they blend into the background by displaying art and making the art look as it is mounted on a white background while also being enclosed in a frame – i.e. like a painting. Samsung does this well and the screens are excellent, but it is limited to art works, leading to the idea that it would be nice to frame websites, specifically ‘data’ based web sites and hang data on walls, data that changes in Realtime. Of course, any screen could be used, but if you simply show data on a screen or monitor, no one notices. If you hang it in a frame, as if its art, people will notice.

The concept was therefore simple, build an application that puts a picture frame around websites, allow users to select which websites to show and if there is more than one, cycle through them, like a dashboard. This would allow small screens, such as iPads, to act as frames and, via AirPlay, devices such as projectors or general screens to gain a look beyond their norm and display websites and data in the same way as hanging art on the wall.

The problem was, this was an idea, to code it would require knowledge of Swift, a library of ‘picture frames’ and probably a business case to justify the time of resident computer scientists (I work at University College London, so we have a few around) and to be honest it was such a simple idea I’m note sure anyone would listen. So, with the rise of AI, I decided to build it myself, using AI from start to finish, from knowing little about Swift, and only as much about AI as my then twitter feed was full of and reading articles on site such as The Verge.

The app took me two days to build, two days where I was also working, so mainly doing things in-between other tasks, over lunch and a little bit in the evenings. With the initial learning curve out the way, I could rebuild it in a day. The first things I did was sign up to OpenAI to gain access to GPT 4 , which released on March 14th 2023. This allowed me access to the latest version and therefore the most up to date knowledge base, although looking back version 3 would probably have been fine and I could have simply used the free tier. My first prompt was “Can you write me swift code to take a website, centre it and add the image of a picture frame around it’. The app should work full screen, in landscape mode”. I had a sample picture frame (taken from Wikipedia and cutting out the actual image to leave a transparent centre).

Within 60 seconds, GPT gave me a section of code and then stopped, it turns out there was a limit in the amount of characters it could respond with. A quick Google search (ironically) showed me that by typing ‘continue’ ChatGPT will continue the code. With this I became a master of cutting and pasting, with the code often taking three continues and code which while sometimes was in code boxes, was also in simply text. It also allowed me to start to understand the layout and nature of Swift, within 30 minutes i had my first app running on an iPad, via Xcode. Simply by cutting and pasting and pressing ‘play’. Not all things worked out, there were often errors, but errors I could cut and paste back into GPT and it would solve them, most of the time.

Each time a milestone was made – a working version, a version with a picture frame in and a website showing, I would save the code as Chat GPT would sometimes break the next version while I was asking for new features. Features such as ‘Add a button to move to the next website’, ‘If there is more than one website, then cycle through them every 60 seconds’ (the websites were hard-coded at this point). Once i had a concept running on my own iPad, I realised it was quite neat and decided to then adapt it so others maybe able to use it. I added a settings page – via ‘Can you delete all the websites I have added and include a settings page, the settings page should be reached via a button and have the ability to add and delete websites, once added they should be saved so the app remembers them when reopened’. This is coding but in the new language – human language.

The picture frames were created in Image Creator by Bing, a search engine I never thought I would use (being a Mac user) but in that last week I had not been near Google. Bing allowed me to type in prompt and get images without any user limits. Simple prompts such as “create me a gold Victorian picture frame, it should be photorealistic with minimum reflections and the centre should be cut out” worked amazingly well, providing me with an almost limitless range of Picture frames to include as part of the assets.

Not all the picture frames are from AI, a couple are from open source imagery, it felt wrong to cut out the picture, but the frames have some provenance and they were nice to include.

Marketing requires imagery as well – as such I took a very simple picture of my two test iPads and used the ‘PhotoRoom” AI app to transform it to two iPads on a bench with a concrete wall behind, whereas the reality was far from as glamourous.

The App is now available via the Apple Store, it even has its own website and logo – again designed by AI, designed using Looka and hosted on Framer. So while it worked, it was a little painful with the limited lengths of code in the early version of ChatGPT and the endless cutting and pasting/fixing things – it was quick, but it still took a notable commitment and time. It is also a very very niche app, but as we get to, thats the point of Vibe Coding.

And then Everything Changed Again

That was 2023 – late 2024, and everything has changed again, not only the workflow and the tools, but also the ability of the LLMs to Vibe Code, they now work, no more cutting and pasting small chunks of code, everything works. Code comes out ready to go and any errors can be fixed in a complete code base, with notably less errors compared to 2024 . As such, we made another app – again Vibe Coding, without knowing at the time – MQTTFrame.

The aim was to build on the concept of FrameIT but allow realtime data intergration directly in the app via MQTT (a lightweight, publish-subscribe network protocol for data) and to make it easy to subscribe to any MQTT topic and display live updates.

Our aim was to allow:

• Customizable Frames: Choose from multiple high-quality decorative frames to complement your home or office décor.

• Clean and Clear Data Display: Showcase important data such as temperature, pressure, news, and more with clear typography and layout.

• Fullscreen Mode: Tap to seamlessly switch to an immersive fullscreen view, maximizing readability and aesthetics.

• Easy Management: Add, edit, and reorder topics quickly through an intuitive interface.

• Instant Updates: Receive real-time data updates over MQTT, ensuring you’re always in the know.

And that’s what we did – but this time in a day and with everything created in ChatGPT – icons, graphics, code, marking text, and even walking me back through how to get it on the Apple Store (as it’s an oddly complex and easily forgotten process).

So one day Vibe Coding and an app was taken from concept to ready to publish on the Apple Store – if you want to try it its currently available for free, on the Apple Store

Everything Again and Again

That was the end of 2024, now in mid 2025 and term Vibe Coding has come into common use and Large Language Models are now more than capable of making anything from webpages to apps to writing posts (all of this was, however, written by an actual human; any grammatical errors are of course, intentional). We have now switched from ChatGPT to Google Gemini and use it almost daily to produce and make apps or webpages that perhaps only I want in my life, such as – an Asteroids Clock, made in 30 minutes (temporarily housed at – https://finchamweather.co.uk/astroidsclock.html – we will move it to Github soon as we get chance).

or a Procedural City Maker (https://finchamweather.co.uk/proceduralcity.html)

or a Weather Dashboard with the forecast made into the image of an English Landscape using data from the API from the UK Met Office:

You can view the dashboard live at https://finchamweather.co.uk/aiweatherimage.html

All small, niche things that somehow I love and I am happy that Vibe Coding allowed them to exist – that’s the magic, Vibe Coding allows you to create anything in your imagination that did not exist to exist, it’s a great time to be making digital things.

If you liked this post, please do subscribe, it gives us a good reason to write another one…

The post Vibe Coding: From iOS Apps to an Asteroid Clock and 3D Cities appeared first on Digital Urban.

pFlow 3ds Max

To achieve more substantial changes in particle properties and behaviour, you can create a flow. The flow sends particles from event to event using tests, which let you wire events together in series. A test can check, for example, whether a particle has passed a certain age, how fast it’s moving, or whether it has collided with a deflector. Particles that pass the test move on to the next event, while those that don’t meet the test criteria remain in the current event, possibly to undergo other tests. The simple example pictured above details a pFlow dialogue determining the birth of particles linked to a target geometry. The particles can subsequently be baked (using pFlow Baker) into an animation timeline for simple output via .fbx, allowing import into external systems such as Unity or Lumion.

The clip above illustrates the pFlow system imported into Lumion with the addition of a scene created in CityEngine.

The post Particles – 3dsMax and Lumion/Unity appeared first on Digital Urban.

The post The Making of the CASA Oculus Rift Urban Roller Coaster appeared first on Digital Urban.

Replicator 2 Printing Cogs

3D Printed Parts

3D Printed Clock

The post 3D Printed Mechanical Clock appeared first on Digital Urban.

1. A three dimensional base model was created with SketchUp on top of the UCL campus map;
2. Moving agents and transports are added into the application to show local bus and tube routes;
3. Sounds are embedded into the application from the UCL Soundcloud;
4. Online data on buildings, Twitter and PC Cluster availability  – live;
5. The user’s location is shown in the application map.

Augmented Reality UCL Map

The post UCL Live Campus Augmented Reality App – Created by Masters Students at CASA appeared first on Digital Urban.

UCL Quad – Lumion

Regular readers will recognise the model from our previous Elder Scrolls IV Game Engine work. We found the model on an old hard drive for a group of students working on a new UCL project. As such we took the opportunity to  load it into Lumion and add it into a future Bloomsbury created using CityEngine. Finally, the falling cubes (CASA branded, home of digital urban) are via 3D Max, linked in from a previous tutorial on Mass FX:

The post UCL Quad – Procedural City and Lumion appeared first on Digital Urban.

Police Car Agents in the City

The post Police Agent City Ball – Unity/NavMesh and CityEngine appeared first on Digital Urban.

Table Top AR City

The post Augmented Reality City Table Top – Unity, CityEngine and Vufoira appeared first on Digital Urban.

Agents in Unity

The post Shortest Path Modelling and NavMesh in Unity and CityEngine appeared first on Digital Urban.