Is no one going to mention that the first picture is AI generated tosh? I understand that generating a picture is easier than taking a nice picture, in a nice place with a a nice camera - but the back row of the board is all white and the second row.. sort of disappears!
All the pictures for shop listing of PCB are also AI generated/modified.
LEDs look completely wonky, the magnet sensors are not there, hole pattern for through hole parts is inconsistent between pictures and none of them match gerber files or video. Amount of tiles on the board vary between 7 and 9. Size of MCU board relative PCB is way bigger than it should be in real life. And of course MCU board itself is also a mess.
It's already bad for the DIY project, but if you are selling something even a single crappy picture would have been better.
I spent far too long trying to work out what was wrong that image, and I was going to say that I didn't like how the bezel ate into the edge squares, until I realised it was AI generated
I think they took the third photo, which is very similar but looks real, and asked an AI to touch it up in some way. Then they probably didn't look very closely at the result.
I think photos are real, just focus is screwed up and maybe they added some automatic effects (tilt-shift or blur afterwards). First few seconds of the video are identical and you can see how much camera is fighting to focus when hand appears in the front.
Yea i hate this trend. I can understand why one would do it for text-only blogs/articles, but for something like this?
The actual product already looks really good, adjust some colors or lighting and that's perfect for a title image
Really?! Making this project must have taken tens or hundreds of hours - if they are proud of the work, why not show it off? The video looks like it wasn't faked.
Did nobody look at the image after they generated it? If they live in a bedsit and don't have a nice windowsill, they could easily generate the background and if the product is vaporware they could drop a rendering over the top - I think that might actually be easier than training an AI on your product... I'll be grumpy at clouds next!
Cute and I'm glad someone is doing that, but it's too small and out of proportion for tournaments etc. You want squares of about 2.25 inches and king height of 3.75 inches, about. If you're going to all the trouble of magnetic sensors and whatnot, you might as well make the board and pieces meet the relevant standards.
These days, image recognition is good enough that it's probably feasible to just video the chess game on a non-sensory board, and let software figure out what moves were played. In cases of doubt or dispute (blitz scrambles), humans can examine the video.
If anyone cares, the tournament sensory sets that most organizers use are made by DGT and cost around $600 iirc. The magnetic sensor system is very clever and was patented in the 1990s or so, but the patents would be expired by now.
The DGT Pegasus [1], which is designed for online play as the device presented her, costs 189 Euro. I know one of the founders of DGT and once applied there for a job for the app development, but did not feel that there was a match at the moment also because I am not really into chess.
That is nice! It's 35cm which is fine for casual games, though online tends to be speed chess and bigger pieces help there. Smaller pieces are fiddlier and easier to fumble if you move too quickly. It would be nice if the Pegasus at least had LEDs at each square to tell you the opponent's move, instead of presumably making you read notation from the screen to transfer the move. But it's a start, and it's fairly affordable.
I do think almost all online competitive players don't use anything like that. They play on screen using a full sized screen and a mouse.
For a DIY project the problem with scaling up is that ordering 50x50cm PCB is going to be quite pricy. Taking into account minimum order quantity of 5, that's ~$300-$600 according to https://pcbshopper.com/ . And that's blank PCB no components or PCBA. If you can find 10-20 people locally to share a bigger batch (good luck with that) you can get it down to more reasonable $25-$40 per board.
To keep this cost effective at DIY level, it needs to be redesigned in a more modular way for larger sizes instead of having one giant PCB. For DIY project might even handwire it using through hole magnet sensors and addressable LED strip.
Surprisingly quote for 5x64 ~= 350 2x2cm boards was as low $30-$90. I knew it would be cheaper but didn't expect that much. That's assuming I didn't break the quote calculator by hitting a weird edge case. Happy middle ground would probably be 2x40cm strips one per row or something similar.
Also the PCB isn't open source. You only get gerber files not the original project files for schematics or board layout which would be needed to modify it.
PCBs get more expensive with larger area, but it should probably be ok.
I hunted down the Gerbers and uploaded to JLCPCB real quick, the cost for the bare PCB at the current size (218x218 mm) is $22 for 5 pieces, excluding shipping. I thought it was interesting to have as a reference.
Scaling up the PCB is not possible in their web UI since the Gerbers "lock in" the size, of course.
Can confirm—I made a smart chess set like this years back by soldering Hall effect sensors and wires manually to a wood board for 64 squares. Every new soldered connection feels exponential—totally makes sense they didn’t do it this way.
I would’ve probably done 4 4x4 PCBs instead so a single damaged PCB could be swapped out.
"I appreciate your work, but that AI generated image is very disturbing and shouldn't be promoting your own work, which is so much more than an AI slab."
I've been thinking about something like that for a while but never got to working on it: having to design, order and wait for a pcb to get fabricated and shipped half way across the world, only to find out that you messed something up is insanely off putting. I so wish this process gets democratized in the future the way 3d printing and laser cutting did ~10 years ago.
Is no one going to mention that the first picture is AI generated tosh? I understand that generating a picture is easier than taking a nice picture, in a nice place with a a nice camera - but the back row of the board is all white and the second row.. sort of disappears!
All the pictures for shop listing of PCB are also AI generated/modified.
LEDs look completely wonky, the magnet sensors are not there, hole pattern for through hole parts is inconsistent between pictures and none of them match gerber files or video. Amount of tiles on the board vary between 7 and 9. Size of MCU board relative PCB is way bigger than it should be in real life. And of course MCU board itself is also a mess.
It's already bad for the DIY project, but if you are selling something even a single crappy picture would have been better.
I spent far too long trying to work out what was wrong that image, and I was going to say that I didn't like how the bezel ate into the edge squares, until I realised it was AI generated
I think they took the third photo, which is very similar but looks real, and asked an AI to touch it up in some way. Then they probably didn't look very closely at the result.
I guess they only have one set of pices, and they used AI to replace them with more fancy models.
They do look weird but they all match a video.
I think photos are real, just focus is screwed up and maybe they added some automatic effects (tilt-shift or blur afterwards). First few seconds of the video are identical and you can see how much camera is fighting to focus when hand appears in the front.
4 replies →
Yea i hate this trend. I can understand why one would do it for text-only blogs/articles, but for something like this? The actual product already looks really good, adjust some colors or lighting and that's perfect for a title image
It's the new normal.
Really?! Making this project must have taken tens or hundreds of hours - if they are proud of the work, why not show it off? The video looks like it wasn't faked.
Did nobody look at the image after they generated it? If they live in a bedsit and don't have a nice windowsill, they could easily generate the background and if the product is vaporware they could drop a rendering over the top - I think that might actually be easier than training an AI on your product... I'll be grumpy at clouds next!
1 reply →
And the knights are messed up
Cute and I'm glad someone is doing that, but it's too small and out of proportion for tournaments etc. You want squares of about 2.25 inches and king height of 3.75 inches, about. If you're going to all the trouble of magnetic sensors and whatnot, you might as well make the board and pieces meet the relevant standards.
These days, image recognition is good enough that it's probably feasible to just video the chess game on a non-sensory board, and let software figure out what moves were played. In cases of doubt or dispute (blitz scrambles), humans can examine the video.
If anyone cares, the tournament sensory sets that most organizers use are made by DGT and cost around $600 iirc. The magnetic sensor system is very clever and was patented in the 1990s or so, but the patents would be expired by now.
The DGT Pegasus [1], which is designed for online play as the device presented her, costs 189 Euro. I know one of the founders of DGT and once applied there for a job for the app development, but did not feel that there was a match at the moment also because I am not really into chess.
[1] https://dgtshop.com/products/chess-boards/dgt-pegasus
That is nice! It's 35cm which is fine for casual games, though online tends to be speed chess and bigger pieces help there. Smaller pieces are fiddlier and easier to fumble if you move too quickly. It would be nice if the Pegasus at least had LEDs at each square to tell you the opponent's move, instead of presumably making you read notation from the screen to transfer the move. But it's a start, and it's fairly affordable.
I do think almost all online competitive players don't use anything like that. They play on screen using a full sized screen and a mouse.
1 reply →
Trivial to scale, since it's open source.
The C++ code is a bit naive, but easily extendable to use a proper engine. https://github.com/Concept-Bytes/Open-Chess/blob/main/Chess....
For a DIY project the problem with scaling up is that ordering 50x50cm PCB is going to be quite pricy. Taking into account minimum order quantity of 5, that's ~$300-$600 according to https://pcbshopper.com/ . And that's blank PCB no components or PCBA. If you can find 10-20 people locally to share a bigger batch (good luck with that) you can get it down to more reasonable $25-$40 per board.
To keep this cost effective at DIY level, it needs to be redesigned in a more modular way for larger sizes instead of having one giant PCB. For DIY project might even handwire it using through hole magnet sensors and addressable LED strip.
Surprisingly quote for 5x64 ~= 350 2x2cm boards was as low $30-$90. I knew it would be cheaper but didn't expect that much. That's assuming I didn't break the quote calculator by hitting a weird edge case. Happy middle ground would probably be 2x40cm strips one per row or something similar.
Also the PCB isn't open source. You only get gerber files not the original project files for schematics or board layout which would be needed to modify it.
2 replies →
PCBs get more expensive with larger area, but it should probably be ok.
I hunted down the Gerbers and uploaded to JLCPCB real quick, the cost for the bare PCB at the current size (218x218 mm) is $22 for 5 pieces, excluding shipping. I thought it was interesting to have as a reference.
Scaling up the PCB is not possible in their web UI since the Gerbers "lock in" the size, of course.
1 reply →
Maybe I’m missing something, but why have a single piece PCB instead of individual parts with a 3D printed slot for the sensor and LED.
A few wires later and a model scale up and your chess board can be a different size.
I suppose one piece is easier and cleaner, just less flexible.
Ordering and wiring up 64 small individual PCBs, vs one PCB, is not a hard choice.
Can confirm—I made a smart chess set like this years back by soldering Hall effect sensors and wires manually to a wood board for 64 squares. Every new soldered connection feels exponential—totally makes sense they didn’t do it this way. I would’ve probably done 4 4x4 PCBs instead so a single damaged PCB could be swapped out.
A comment from the article:
"I appreciate your work, but that AI generated image is very disturbing and shouldn't be promoting your own work, which is so much more than an AI slab."
I've been thinking about something like that for a while but never got to working on it: having to design, order and wait for a pcb to get fabricated and shipped half way across the world, only to find out that you messed something up is insanely off putting. I so wish this process gets democratized in the future the way 3d printing and laser cutting did ~10 years ago.
You can use a cheap desktop CNC machine to mill circuit boards.
PCBs here: https://www.patreon.com/c/ConceptBytes/posts