Memory and speed¶

Image models are large — from about 1 GB to over 10 GB — and a phone has 8 to 16 GB of RAM for everything. Nightmare manages this for you; this page explains what it does, so a slow first Run or a model being let go is not a surprise.

A loaded model stays loaded — while you are using the app¶

The first Run after choosing a model loads it, which takes a few seconds (longer for big models). After that it stays loaded, so the next Run starts at once. The line under the top bar shows it: AbsoluteReality (idle) means loaded and waiting.

Switching to a different model — or, for SD 1.5, a different resolution — means loading again. A flow that uses two models loads each once per Run, and the app tells you before you press Run.

Models too big to keep¶

Some models take so much of the phone's memory that a second render in a row would not fit. For those, the model is let go after each Run, and the next Run loads it again (a few seconds, since the phone keeps the files cached). The run log says so when it happens.

On a 12 GB phone this applies to Z-Image Turbo and Qwen Image 2.1. On 16 GB phones they stay loaded like the others.

Qwen Image 2.1 also loads its parts one at a time — the text encoder, then the image model, then the decoder — giving each back before the next. That is why it fits on a 12 GB phone at all, and part of why it is slower.

When you leave the app¶

Close the app — swipe it away from recent apps, or back out of it — and the model is let go at once, even in the middle of a render (a closed app has abandoned it).

Step away — Home, switching to another app, the photo picker — and the model stays for one minute, so a quick trip does not cost a reload; after that it is let go, or at once if the phone reports it is short of memory. A render you step away from is never interrupted: it finishes first, then the minute starts.

Coming back costs one reload on your next Run.

Low RAM mode (SDXL and Anima)¶

SDXL and Anima models can either stay fully loaded (faster, with a live preview while they render) or load each part on demand and free it after use (low RAM mode). Below 16 GB of RAM, low RAM mode is required.

Settings → General → Memory has a switch for each, set automatically from your phone's RAM. You only need to touch them if:

you have 16 GB and an SDXL or Anima model is killed during a render → turn low RAM mode on;
you have 12 GB and Anima is still killed with low RAM mode on → also turn on Anima DiT sequential loading (slower per step).

If the app is killed during a render¶

Android closes apps that use too much memory. If a render stops and the app restarts:

Close other apps, especially games and camera apps, and try again.
Use a smaller size — memory grows with the picture's area.
For SDXL or Anima, check low RAM mode is on (above).
For a FLUX.2 edit with a reference picture, use 512 × 512.