gyuforge (@gyuforge) header
gyuforge (@gyuforge) profile picture

Solo developer building games and software. I share development problems, useful tools, and practical AI experiments.

Joined October 2026
Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)Photo by gyuforge (@gyuforge)
Move photos from an iPhone to a Windows PC without emailing yourself a copy. LocalSend transfers files between devices on the same local network. Install it on both, choose a file and pick the receiver. Free and open source. No account or cloud upload. localsend.org/
28
ChatGPT just saved me from buying a new router lol. Seven appliances on 2.4 GHz kept dropping off Wi-Fi every 30 seconds. Even the 5 GHz connection I normally used would cut out completely at times. Firmware updates and channel changes hadn't helped. After going through 5,000+ log entries with ChatGPT, we found wireless multi-bridge still enabled. I'd set it up to extend the network, stopped using the extra router, and completely forgotten about that setting. It wasn't even connected to an upstream router. After turning it off, disconnects among the 2.4 GHz appliances went from 26 to zero in matching 2-minute checks. Without ChatGPT, I would probably have written it off as a hardware failure and replaced the whole router. If your Wi-Fi keeps dropping, check for forgotten extender or bridge settings before buying a replacement. Ask ChatGPT to check the logs, official documentation and similar cases to find suspect settings and help fix them.
1
1
61
Inspect Windows event logs in your browser. DevToolLab’s EVTX Viewer opens .evtx files and lets you filter by event ID, inspect individual fields, and read the XML. Its built-in sample has eight failed logons under event 4625. Open one and you can see the wrong-password code with a plain-language explanation. Free, no account required. devtoollab.com/tools/evtx-vi…
37
A debugging tool should tell you when it has dropped part of the log. Otherwise a normal-looking answer can be based on a different input from the one you sent. If a long log seems to be ignored, check the runner’s truncation warnings before switching models.
Ollama gives a local model 4,096 tokens of context on GPUs under 24 GiB. Sent an 18,041-token log, its llama.cpp runner kept the first 4 and last 2,046 tokens and returned no error. OLLAMA_CONTEXT_LENGTH=64000 fixes it: devtoollab.com/blog/how-to-r…
25
Cutfac saves the before and after of an agent’s video edit so you can compare them and restore the earlier version. The workspace is ordinary folders and YAML.
I can't use CapCut. I still made 300 short videos with Claude Code, and they got millions of views. The hard part was never the first draft. It was fixing it. So I built an editor with no editing tools: box what's off, type the change. Open source, Mac.
35
A Linux desktop can boot in a cloud workspace and still be painfully slow. The QEMU TCG line in this screenshot is worth checking. TCG is QEMU’s software CPU emulation mode. CPU virtualization and desktop rendering are separate things to check. Before tuning animations, check whether hardware virtualization is available to QEMU and whether it’s actually enabled.
Replying to @Replit
@Replit can do anything. Tell me another platform that can do this. Running Omarchy inside workspace. This is insane. (Super super laggy though. trying to optimize)
27
Trimming one scene out of a long gameplay recording shouldn’t mean re-encoding the footage. LosslessCut starts cuts at keyframes and copies the original data. In our test, a 59-second video became a roughly 7-second clip, with both video and audio preserved without re-encoding. If your game audio and microphone were recorded as separate tracks, you can also remove the mic track while keeping the video. Free downloads for Windows, macOS and Linux: github.com/mifi/lossless-cut
2
3
68
RTX 5090 and 5080 listings are missing from NVIDIA's US marketplace. They're still listed on the Korean marketplace. The RTX 5070 Ti launched with a $749 starting MSRP. One ASUS ROG Strix listing shown here asks $1,499.99. The Korean retailer listings in these screenshots work out to roughly $1,490–$1,860 for the RTX 5070 Ti and $8,760–$9,970 for the RTX 5090. High-end GPU availability and pricing are concerning. NVIDIA has denied that the RTX 5090 is being discontinued, though Jeff Fisher acknowledged supply constraints on October 16. Where is the gaming GPU market heading? Will prices just keep climbing?
79
Same photo saved under different names and sizes? Krokiet can find identical files by content, plus visually similar images such as resized copies. Free and open source for Windows, macOS and Linux. Preview matches before deleting. Similar-image matches can be wrong. github.com/qarmin/czkawka/tr…
46
Throw some marbles and blow off some steam! RUMBLORB: Marble Mayhem is available on Google Play.
3
54
Here's a teaser for Lumenfold. I've been working on story cinematics and boss battle videos, and I'm continuing to refine the gameplay. Thanks for checking it out.
6
50
The model is done. Now it needs to move—and you still need a demo that shows it clearly. Three free tools for animating models, generating motion, and polishing demo recordings: • Mesh2Motion • NVIDIA Kimodo • OpenScreen A look at each:
1
1
57
2. NVIDIA Kimodo Generate human motion from text, with paths and pose constraints to guide it. SOMA models can export BVH. Runs locally. Up to 10 seconds per prompt; retargeting and cleanup are still part of the job.
1
20
3. OpenScreen Turn a screen recording into a clearer demo with click-based zooms and silence removal. In recordings with separate cursor data, you can adjust the cursor afterward. Free and open source on Windows, macOS and Linux.
24
You remember the error, but not where you saved it. Kensaku lets you: • Find error messages inside screenshots • Search text in PDFs and logs • Find documents without the filename • Let AI agents search local files via MCP Free, local, open source. github.com/yjid/kensaku
1
42
Long conversations in ChatGPT dots have become so sluggish that even typing is difficult. Separately, in Codex, I’ve kept running into “unexpected status 413 Payload Too Large: Request too large” errors lately and having to start fresh conversations. It feels like that eats into the benefit of cache hits.
1
75
I shipped on Google Play a while ago, but getting onto the App Store still feels like a much higher hurdle.
1
3
178