Performance Monitoring
When a generation takes twice as long as expected, or a run fails with an out-of-memory error, the problem usually isn't your prompt — it's your hardware hitting a limit. Performance monitoring helps you see what's happening under the hood so you can separate creative decisions from system bottlenecks.
What to Watch
Four resources drive generation performance. When one of them is constrained, you'll feel it:
| Resource | What it affects | Warning signs |
|---|---|---|
| GPU utilization | How hard the graphics card is working | Low utilization during generation = a different bottleneck is slowing things down |
| VRAM | How much model/image data fits on the GPU | OOM (out of memory) errors, crashes during generation, forced restarts |
| CPU | Preprocessing, model loading, I/O operations | Slow model loads, sluggish UI responsiveness |
| Disk | Reading models, writing outputs, project file access | Long pauses before/after generation, save operation lag |
Performance Troubleshooting
When something feels slow or is failing, this table helps you zero in on the cause:
| Symptom | Likely bottleneck | What to try |
|---|---|---|
| Runs much slower than expected | Model too large for GPU, or thermal throttling | Check GPU temp and utilization; try a smaller model |
| Frequent OOM errors | Not enough VRAM for the resolution/model/batch combo | Lower resolution, reduce batch size, or use a lighter model |
| Queue feels stuck with low GPU usage | Engine/dependency issue, not a GPU problem | Check engine logs and dependency health |
| Long pauses before generation starts | Large model loading from slow storage | Move models to SSD; check disk I/O |
| Save operations lag heavily | Disk I/O bottleneck on output directory | Verify free space and storage speed |
| UI becomes unresponsive during generation | CPU or memory pressure | Close other heavy applications; check system RAM |
Status Bar Interpretation
The status bar at the bottom of Studio provides quick signals about system health:
- Engine warnings — Connection or runtime configuration issues. Address these before anything else.
- Resource warnings — Storage pressure, I/O issues, or memory constraints.
- Dependency warnings — Missing models or nodes that could affect the next generation.
Treat these as early-warning signals. A resource warning during generation usually means you're close to a hard failure — it's better to address it now than to wait for the crash.
Performance HUD
The Performance HUD provides a real-time overlay of system resource usage during generation. When enabled, it shows GPU utilization, VRAM usage, CPU load, and other runtime metrics while you work.
This is useful when you're trying to understand why a specific Pipe or model combination is behaving differently than expected. Watch the HUD during a generation to see exactly where resources peak and where bottlenecks appear.
Tip: You don't need the Performance HUD running all the time — it's a diagnostic tool. Turn it on when investigating a performance issue, then turn it off once you've identified the problem.
Diagnostics Bundle
When you need to capture a comprehensive snapshot of your system state — for your own records or to share with support:
- Open the diagnostics section in Settings.
- Export a diagnostics bundle. This collects system info, engine status, dependency state, and recent logs into a single package.
- Save the bundle file. It's useful for comparing "before" and "after" states when troubleshooting, and it's the fastest way to give someone else the context they need to help you.
Patch Notes and Version Checks
When generation behavior changes unexpectedly, check whether a recent update could be the cause:
- Open About or Patch Notes in Settings to see what changed in the latest version.
- If a performance regression appeared after an update, the patch notes may explain new requirements or known issues.
- The version display also shows your build and channel, which helps when comparing notes with other users or reporting issues.
Note: If an update indicator appears during an active generation session, finish your current work before updating. Don't update mid-session — complete what you're doing, sync if needed, then update.