Q everyone!

I’m Filatov Anton, a web architect specializing in e-commerce development.

But this channel is more about other things; here’s what you’ll find:

• DIY engineering: 3D printing, CAD, electronics, and smart home
• Unconventional technologies for project implementation
• Live updates on current projects
• Experiments with AI and various aspects of coding

Where else to find me: GitHub · Thingiverse · YouTube

And, of course, check out my blog – positroid.tech

So, why do I need this channel?

I have many mini-projects, interesting ideas, or useful software. It’s not enough for a full post, even for my blog. I’m not even talking about other platforms.

Here, I also share current progress with projects and interesting solutions that often get lost when I try to write a full article.

Also, sometimes I want to share something in the moment while my emotions and feelings are still fresh. The next post will be about that.

Think of this channel as an addition to my blog and YouTube videos. It will also have announcements for posts and videos because it’s difficult to follow the blog (there’s only an RSS feed to subscribe to).

I decided to make good videos for YouTube, with 4K, high-quality images, audio, and all that.

A 4K display is coming to me (I use two FHD monitors — it's good, but I need 4K for screencasts). I'm also getting a Raspberry Pi 4B instead of my dead Orange Pi. I need it for Klipper with 4K timelapses of 3D prints.

That's a bit of a tangent. I needed a logo and an intro for YouTube. I made the logo years ago when the first image generation AIs appeared. It was Stable Diffusion, which couldn't even reliably generate a single letter "P". I generated the intro video only now.

I used Kling, inserted the logo image, played around with prompts for animation, and generated the audio for it. And voilà — the ready 5-second intro!

If I did this manually, I would need to model everything in 3D, learn After Effects, and spend at least a couple of weeks on it — and all that without audio. In other words, I really like modern technologies and the AI boom. It has radically decreased the entry threshold in almost any area.

Here is one of the first 4K print timelapses for one of the upcoming projects that prompted me to upgrade to a Raspberry Pi 4B.

Main reasons and benefits of switching to Klipper:

1. After replacing the extruder from a BMG to a Biqu H2 V2S, the printer could no longer handle high-speed printing and would violently vibrate at minimal direction changes at high speed.

2. No more running back and forth with a microSD card between the printer and the PC — now files are uploaded by dragging and dropping into Fluidd.

3. Finally, genuine timelapses with good image quality and settings.

By the way, a life hack for those who don’t want to buy a separate camera for timelapses — you can use an Android smartphone for this purpose by installing the IP Webcam app. The app starts a server that streams the phone’s camera feed — it offers tons of settings and multiple streaming protocols, including snapshots.

There will be a full blog post about installing Klipper on the Sapphire. Consider this an announcement.

The promised article about implementing a Home Assistant plugin for managing the ESP32 interface is available on the blog.

One might think it was just a matter of porting an existing interface into HA—what could possibly go wrong? As it turns out, everything can go wrong.

Details are in the article; here are a couple of takeaways:

AI knows how to write Python plugins for HA, but gets tripped up on the details;

The HA authorization mechanism is nontrivial—via WebSocket you must obtain a signed key for each endpoint and append it to the URL—in my case, virtually every request to the robot API goes through this mechanism.

This series of articles about the Caretaker is almost complete—there will still be a video showing the manufacturing process, fixing a couple of HA-specific bugs, and small enhancements (in the YouTube comments they requested WASD controls). I’ll probably fit all that into one more article in due course.

New device added to the arsenal!

You probably wouldn’t go out of your way to look up what a cat looks like through a thermal camera, so here’s mine.

P.S. This is the Mileseey TR256i — I’m not a professional, so in cases like this I choose budget-friendly devices with decent reviews and specifications.

I've got a bit more to say about Klipper and timelapses – particularly observant people might have noticed dropped frames.

There’s an issue on GitHub reporting the same problem with a similar stack – people were also using IP Webcam on a phone as the camera source.

This led me to the idea that the root cause lies in the timeout for downloading the snapshot image.

I had a bit of free time and tried to dig into the error myself. I cloned the repository, opened it in Cursor (an IDE with AI), pasted the text from the issue, and asked it to fix the problem. The first prompt already gave me working changes. It turned out to be trivial — the timeout was hardcoded to 2 seconds, so I added a configurable parameter for it.

I set it to 5 seconds and did a test print with 80 layers — not a single frame was dropped.

The issue is resolved now, so I submitted a Pull Request to the repository. Unfortunately, I doubt it’ll get merged, the project seems to have been abandoned for a couple of years now.