Log inCityLife v0.2.8 β now it runs without an NVIDIA graphics card π
Until now, CityLife's AI image engine assumed a modern NVIDIA GPU, which left a lot of laptops and older PCs out. That's over. Version 0.2.8 can generate portraits, scenes and selfies using only your processor β so if your PC has integrated graphics, or an old graphics card, you can finally see your city's people.
β¨ What's new
- π₯οΈ Image generation on the CPU. No NVIDIA card needed. The SD1.5 engine runs on the processor and picks the right device by itself.
- ποΈ New "Device" setting in the image settings: Auto, GPU or CPU β with a short note explaining why the game chose what it chose.
- π‘οΈ Safer on weak graphics cards. If your GPU has less than 4 GB of VRAM, or a driver too old for the AI engine, the game now falls back to the CPU instead of crashing.
- π Automatic switch to SD1.5 on PCs without a usable NVIDIA GPU (SDXL still needs one, and now says so clearly instead of failing).
- π¬ Errors that actually say something. If the AI engine stops in the middle of an image, the game now shows what happened and where to find the log β not just a cryptic network error.
- π Tidier start page. The Donate, Patreon and SubscribeStar buttons are now stacked in a clean vertical column.
β±οΈ What to expect on the CPU
It works, and the quality is the same β but it's slow. Plan for a few minutes per image, and keep around 4 GB of free RAM (close the browser and other heavy programs before generating). The game waits up to 30 minutes per image in this mode, so it won't give up on you.
Tip: generate a few images, then go back to playing while the engine works.
π The story behind the fix
This update came from a real test machine: an old laptop with a GeForce 940MX (2 GB) and 12 GB of RAM. The AI model loaded fine⦠and then the engine silently died on the very first image.
The obvious suspect was running out of memory, so I trimmed memory use first (0.2.5). It didn't help β but the engine's new log did. The real culprit was a single call: while reading your prompt, the text library asks the graphics driver "is a CUDA stream being captured?" β even when running on the CPU. With a driver too old for PyTorch, that one question crashes the entire process.
The engine now answers that question itself whenever CUDA isn't usable. Result: it works on that laptop. π
π¦ Update notes
- Just download the new portable .exe and run it. Your saves stay in the data folder next to it.
- The first image after starting is the slowest, because the model has to load.
- If something still goes wrong, the error now includes the log path (ai-engine.log) β send me the last lines and I'll take a look.
