https://codeberg.org/blast/blastqr
Hello! I’ve been working on blastQR for a couple of years now and I thought this community might find it useful. Feel free to try it and ask me anything!
blastQR is a fully open-source file transfer app and protocol that works completely offline. You can even use it in airplane mode. Data is transferred through sequences of QR codes displayed on a sender device and scanned by a receiver device’s camera. It uses RaptorQ codes so the receiver can recover the file from any sufficiently large subset of frames.
Note that although I have made use of AI as a coding assistant I do not consider this to be a vibe-coded app and I do plan to continue improving it and supporting it in the future.
See video of blastQR in action: https://github.com/user-attachments/assets/038a2567-70c8-47c8-b165-dea858d1aef3
Here are some of the key features:
- Works fully offline: No internet, Wi-Fi, Bluetooth or servers. Your data stays offline.
- One-to-many distribution: Share the same file with multiple devices simultaneously using a single blastQR broadcast.
- Cross-platform: Android, iOS, macOS, Linux and Windows, plus a command-line tool that can send from a terminal, even over SSH.
- Password encryption: You can optionally encrypt files with an auto-generated password or a custom one.
- Media compression: The app supports lossy image compression and video transcoding for quicker transfers.
- Multiple codes on screen: Shows multiple QR codes at once for higher speed. With 8 codes at 60 fps, transfers peak at around 1.4 MB/s.
- Send multiple files in one go: They’re zipped into one transfer and unpacked into separate files on the receiver.
- Sound feedback: The receiver plays a short tone when it has the whole file, and the sender stops automatically when it hears it.
- Start scanning at any point: No need to catch every code, just keep the camera on the screen until it’s done. Blurry or missed codes don’t matter.
- Regular QR scanner: You can also scan normal QR codes and save them.
- Open protocol that other apps can implement
- Built-in file manager
I’m trying to build a small community around the project so it can be improved and maintained and so it can reach more people in the future.
Downloads for all platforms can be found here: https://codeberg.org/blast/blastqr/releases/latest
I’ve been enjoying this
Seems finicky. Keeping the camera and screen aligned during the transfer. Someone should research if it’s possible for two phones to send data acoustically. Then they don’t need to be kept perfectly aligned. Hell, they could even transfer data over a phone call or something.
I do miss the demon screaming of modems.
A very interesting take on a new malware distribution method.
Does this mean an IP cam pointing to a monitor can compromise an air gap? I guess it always could, just at 1.4mbps now
Dammit you beat me to it especially if you make it nfc capable too. Wink wink hint hint
Note that although I have made use of AI as a coding assistant I do not consider this to be a vibe-coded app
-
What KIND of AI did you use? Was it a local model or a cloud sourced one?
-
Where exactly and what exactly was the AI used on?
OP answered it here:
Not sure if I’m okay with it, then. Cloud models tend to suck up data, so they probably have some kind of idea of how it works. This is why I tend to only be okay with stuff that was vibe-coded with Local LLMs.
It’s open source. Anyone who is curious can find out exactly how it works.
That’s valid, I simply appreciate that OP has been forthcoming about the usage and how it was applied.
-
So a fun use case you may not have considered is back country GPX sharing. A plugin for Osmand (a very popular FOSS mapping app for back country use) would be neat. Sharing when airplane mode between devices is always needed.
That is a fenomenal idea! I hadn’t thought about it and these are the exact kind of situations it could solve well. I’ll definitely add it to the roadmap. Thank you
Reminds of removedat, odd but apparently used in india because their goverment is lowkey evil again?
Bit chat got Scunthorped
An interesting name _removed_at is /j
I started working on something like this years ago, but I didn’t have the coding skills to pull it off. This is really awesome.
Neither does he.
Yeah, we get it, he used AI. You anti-AI absolutists aren’t actually changing anyone’s minds with this behavior, and actually seem to be annoying a lot of people in to the opposite direction.
It’s not that hard. Loop the QR, index the frames without the QR payload, do it ~10fps so you account for camera error, then when all frames/packets are completed, you’re good. There’s a ton of QR scanning libraries out there already.
Is this the future version of holding two phones together to transmit files via infrared?
I don’t see a lot of use cases where blastQR beats LocalSend. The ones I do see are mostly about surreptitious data exfiltration against the will of device and/or data owners, and those are still complicated by the need to install your client on the sending side.
Could be for airgapped devices
Sure, that’s covered by my commentary about data exfiltration against the wishes of data or device owners, and about the difficulty of installing the client on the airgapped device.
I believe there are a few legitimate use cases that don’t have a faster/better alternative, but they’re very niche.
If you’re in an environment where turning on your cell radios could compromise you then you’d like something that operates through something other than radio. Two people sharing data at a protest for example.
LocalSend uses wifi rather than cell but you’re right that a protest could be a good place not to turn on any radios. It’s also a good place not to bring a phone at all, even if you’ve thought in advance to install a visual data transfer program like this. But you have a good point.
There’s a wide range of threat levels someone could be managing at a protest. Lots of people don’t feel particularly at risk from being grabbed and having their phone cracked, but still don’t want to be added into a database of people opposed to a regime because an automated mass surveillance system sniffed their radio signature. If you’re not expecting a personal assault phones can be useful for documenting police, promoting the event afterward, and accessing emergency services.
I don’t think the “don’t bring a phone to a protest” advice is good unless you think it’s likely you could be in a position to lose control of it. The tradeoffs for that level of safety don’t make it worth doing for random citizen protesters. I do think concealing your face and turning off your radios is a good idea, because those are very realistic threat vectors for a generic citizen protester. Of course you could have the best of both worlds by getting a burner phone, but I don’t think many people are that prepared.
Funfact: localsend works between two smartphones if on is on hotspot mod
A couple of months ago I searched for a tool like this, there are a few but all with some flaws: the most polished one did not cycle the codes, the most usable one failed to initiate file download from web UI they had. Have you researched existing implementations?
Also, maybe an app is better for stability, but have you thought about having a web UI, too, so that one can use it to receive or send the file (the page itself would be pretty lightweight, considering it doesn’t send or receive the data itself)? Such a UI could also be served from localhost
Yes, over the years I’ve tried many solutions. The latest popular one being Decimen and others mentioned in the readme.
Since it seems you are interested in the field I’d be very happy if you tried it and told me how it compares to the rest. blastQR is still a work in progress so any feedback would be very helpful!
A web UI is also possible. I’m not working on it right now but it’s something the project should support in the future.
Since you advertise it for Obtainium, would you mind adding the key for verification with AppVerifier? I didn’t find it in the readme and release.
Thank you for the suggestion!
dev.blastapps.blastqr 60:E3:A6:C0:63:1D:C3:A0:0C:78:E4:C9:E4:89:B3:03:4F:A9:7D:49:D0:B9:3E:CF:56:B2:33:A8:C3:DD:F6:83it’s been added to the readme as well.
Dope, thank you very much!
Two more questions: Is there a way to set 8 codes for sending in the flatpak version, or is that limited to 6? Any idea why a pixel 8 pro on graphene wouldn’t allow 120/240 fps receiving? Highest I’ve been able to choose and test is 60fps. was thinking that it may be the resolution cap, but an iPhone comparison ran at 240fps with the same res (I think).
The app is limited to 6 right now but I’ll work on increasing it! The CLI tool does allows up to 12. Make sure you use Ghostty or Kitty for high frame rates if you do decide to try it.
A lot of phones cap the camera preview at 30fps actually so 60fps is quite good already. From the devices I tested only the iPhone let me select 60,120,240. You can see some benchmarks here: https://codeberg.org/blast/blastqr#real-world-benchmarks I don’t think you are missing out on performance if you can scan at 60fps. I recall at 120 the cpu was starting to be the bottleneck but I’ll have to do some more testing to confirm.
Yeah, thanks for the input. i did realise that the iPhone too performs best at 60FPS actually(anecdotally).
This is truly a cool project. thanks for your work!
Isn’t app verifier a dead project?
Note that although I have made use of AI as a coding assistant this is not a vibe-coded app
Can you elaborate more on how AI was used?
For context, I’m a software developer with around 15 years of experience. Most of it writing Android and iOS apps in Java, Kotlin, Swift and lately Dart (Flutter) + Rust which is what this project is written in.
Although I did start the project from scratch I have lately been using Claude Code as a coding assistant. I write code myself but I also let it write or refactor parts of it. The CLI tool is the part that Claude has mostly written from scratch because it’s not something I had experience in writing before. That being said I also review all the code that Claude writes.
Thanks for asking! there’s also some disclosure about this in the readme file
Thanks for being honest and open about it
The CLI tool is the part that Claude has mostly written from scratch because it’s not something I had experience in writing before.
This seems like the perfect opportunity to learn something new in my opinion, CLIs aren’t difficult to make. And you may want to consider moving to GitHub because of Codeberg’s ToS.
I totally agree and I’m very open to welcome anyone who wants to rewrite that part or I might go for it myself in the future. The reason I went for it is that I really wanted to be able to use the CLI tool right away and the AI produced a good result so I left it in the project for now.
About Codeberg I did read their blog post and I didn’t think my project falls into any of the cases they mention:
spoiler
Projects that are created by LLM “agents” in autonomous ways Projects written and maintained with heavy use of LLMs Projects where the amount of resources (e.g. storage, CI/CD) is significantly larger than what the involved amount of people could have created by hand Projects heavily tied to the LLM ecosystem, e.g. LLM-written tools to ease LLM usage Users sending LLM contributions in violation of project’s custom policies
If they ever consider it infringing their ToS I could move it to GitHub where there’s already a mirror of the repo.
What was your testing environment?
First thing I thought about is how well smartphone can handle it. After fps mismatch that you contributed on, there is also cameras matrix’ resolution, probable obstacles and image distortion.
After that, I’m interested in how able a smartphone to scan a max res QR like in the OP gif. Or eight of them. 25 times a second, with a CPU overhead of error correction algorhitm. I know modern phones are beasts, but basic apps I used struggled with QRs that vere too busy, and even if not, they still took a second before popping up a link.
Btw, do you use standard QR by specs, or a custom 2d encoding method? I don’t see rectangular orientation marks on your images, so it’s something new. Since you are not sticking to QRs, have you thought of using colors, since your usecase is capturing color displays?
You bring up great points!
Here are some benchmarks including videos (Note that these are best case conditions but should be achievable): https://codeberg.org/blast/blastqr#real-world-benchmarks
Most of the testing I did with my now 7 years old Galaxy S10e. It only scans at 30FPS but it can handle 6 codes at once which is a pretty decent 420 kB/s. During testing on the iPhone 17 which can scan at 60 or 120fps I think the CPU did not become the bottleneck until around 8 codes at more than 90fps. That is something I still need to test and document more thoroughly though.
I do use standard QR codes. By default Version 40 which can store almost 3 KB. It is configurable in the app’s settings.
Regarding color, libcimbar (https://github.com/sz3/libcimbar) is a project that does exactly that although the speeds are not quite as high yet. Additionally, one of the cpu optimizations in blastQR (and qr scanners in general) is to drop color so that each frame that gets copied in memory is smaller. It makes a difference when you’re processing at high fps.
