Nightmare user guide¶
Nightmare makes pictures and short videos on your phone, with no internet and no account. Everything runs on the Snapdragon's NPU (its AI chip). You describe what you want, press Run, and the picture appears on the screen a few seconds to a few minutes later, depending on the model.
It works a little differently from most apps: every job is a small flow of boxes (nodes) joined by lines (wires): a prompt box, a generator, an output. You never have to build one from scratch: the app ships ready-made flows for every job, and you can just open one and press Run. When you want more control, the same boxes let you see and change every step.
Download the APK · Demo video · Telegram group · All releases · Everything it does · Report a problem
This guide is a work in progress
It covers everything the app does today, but some pages are still thin and the screenshots are drawn by the app itself rather than taken from a real phone, so pictures show as plain colour blocks. Corrections and requests are welcome on GitHub.
Where to start¶
- New here? Start with Install and check your phone, then Your first picture. Ten minutes, most of it the download.
- Want a specific job? Flows has a page for each: edit a photo, redo one area, enlarge, make a video.
- Looking up a setting? The Reference explains every node, knob and Settings switch.
- Something went wrong? Troubleshooting covers the common failures.
What it can do¶
| job | flow | page |
|---|---|---|
| A picture from a description | Text to image | Text to image |
| Re-imagine a photo | Image to image | Image to image |
| Redo one area of a photo, or extend it | Inpaint | Inpaint |
| Paste an object from another photo, blended in | Inpaint + Add Objects | Add Objects |
| Make a photo bigger and sharper | Upscale a photo | Upscale |
| Change a photo by instruction ("make it winter") | Image edit | Image edit |
| A two-second clip from text or a photo | Text / Image to video | Video |
| A better clip of up to 5 seconds | Text / Image to video, Wan 2.2 | Video |
| Try many seeds or settings at once | any flow + Batch | Batching |
What your phone needs¶
- Android 12 or newer, and a Snapdragon processor. Other chips (Exynos, Tensor, MediaTek) have no supported NPU: on one, the app says so and offers no downloads. Text to image and Image to image still run there with an imported SDXL MNN or SD 1.5 CPU model, on the GPU or CPU.
- Which models you can use depends on the chip:
| chip | can run |
|---|---|
| Snapdragon 888 and newer | SD 1.5 models, upscaling |
| Snapdragon 8 Gen 1, 8+ Gen 1 | + Anima DiT (Turbo 1.1 and 2.9B), run on the CPU because the NPU there never finishes a DiT step: about 15 to 20 minutes at 512 |
| Snapdragon 8 Gen 2, 8s Gen 3 and newer | + Z-Image, NitroFusion 2, FLUX.2 Klein 4B and 9B, Anima DiT, Iris (on the 8s Gen 4 and 8s Gen 3 with the NPU's matrix unit off: right pictures, but much slower; Iris is not offered there, it gives noise) |
| Snapdragon 8 Gen 3 and newer | + SDXL and Anima, FLUX.2, Qwen Image, Anima DiT, Iris, Z-Image, MOSS-Image, FLUX.3 Klein, Krea 2 and Fibo 2.0 Edit (12 GB) |
| Snapdragon 8 Elite and newer | + video; Ideogram 4 and Krea 2 with 12 GB of RAM or more |
| Snapdragon 8 Elite Gen 6 | the NPU and DiT models above |
- Storage: models are downloaded separately, from about 1 GB (an SD 1.5 model) to 13.8 GB (Qwen Image on a 16 GB phone). The app tells you each size before you download.
- Memory: 8 GB of RAM runs the SD 1.5 models; 12 GB runs everything, some of it slowly (FLUX.2 Klein 9B starts at 768 to stay inside it; Krea 2 renders in parts, about 3.5 to 4 minutes at 1024 × 1024).
The app checks your chip itself and says plainly when a model or flow cannot run on it.
Privacy¶
Nothing you type or make leaves the phone. There is no account, no analytics and no upload. The app only goes online to download a model you asked for, and a picture only leaves when you share it yourself.