Hey folks! Ran into this super frustrating bug last night where Automatic1111 was just spitting out completely black images, even on simple prompts that worked before. It felt like my GPU decided to take a nap 😅. After some digging and a few Discord chats, I found a couple of fixes that worked for me and my teammates.
The most common culprit seems to be **VRAM running out during processing**. Here's what you can try, in order:
* **Use `--medvram` or `--lowvram` command line args:** If you're on a card with 8GB or less, this is often the fix. Edit your `webui-user.bat` (Windows) or `webui-user.sh` (Linux/Mac) and add one of these flags to the `COMMANDLINE_ARGS` line.
```bash
# Example for webui-user.sh on Linux
export COMMANDLINE_ARGS="--medvram --listen"
```
* **Disable and re-enable the "Add random noise to latent space" setting:** In Settings > Stable Diffusion, uncheck it, apply, then re-check and apply again. Weird, but it cleared it up for some.
* **Check your `clip_skip` value:** If you're using a custom model, try setting `clip_skip` back to 1 (in Settings > Stable Diffusion) to see if it's a model-specific issue.
* **Rollback to a previous commit:** Sometimes an update introduces bugs. You can try:
```bash
git checkout
./webui.sh
```
For me, the `--medvram` flag did the trick instantly. It was a clear sign my other monitoring dashboards were about to spike in GPU memory usage! Has anyone else found a different solution? Sharing your setup (OS, GPU, VRAM) would help others troubleshoot.
Dashboards or it didn't happen.