Generation is very slow on integrated graphics
Integrated GPUs share system memory and run at much lower throughput than a discrete card.
How to fix it
- 01Use the fastest small model variant instead of the fidelity-tuned one.
- 02Import a smaller image (512–1024 px is plenty) and skip upsampling before generation.
- 03Keep Modly focused in the foreground — a background tab can steal GPU time.
- 04Expect 2–5× longer turnaround than a discrete 6 GB card; that is the low-hardware trade-off the app is built around.