NOX + LUX
Image & video
generation.
Lux is the desktop workspace for Nox, a CUDA engine for image and video generation.
Made with Nox
Krea 2 → LTX 2.5
MADE WITH NOX
Every image below was generated with Nox using Krea2.
LUX WORKSPACE
Generate images and video, adjust prompts and LoRAs, and save settings as snapshots.


Lux desktop frontend · Nox inference engine
NOX ENGINE
C++20 and CUDA inference for diffusion models on NVIDIA GPUs, with planned memory use and model offloading.
Warm-run generation time
Lower is faster · seconds| Pipeline | ComfyUI | Nox | Speedup |
|---|---|---|---|
| Anima | 4.81s | 1.73s | 2.78× |
| FLUX.2 Klein 4B | 3.74s | 1.07s | 3.50× |
| FLUX.2 Klein 9B | 6.86s | 1.73s | 3.97× |
| Ideogram 4 | 13.36s | 5.24s | 2.55× |
| Krea2 | 19.61s | 5.16s | 3.80× |
| LTX 2.3 | 46.35s | 12.19s | 3.80× |
| LTX 2.5 | 46.10s | 12.77s | 3.61× |
| MiniMax H3 | 35.67s | 13.53s | 2.64× |
| Qwen Image 2512 | 9.50s | 2.82s | 3.37× |
| Qwen Image 2.1 | 34.33s | 8.72s | 3.94× |
| SDXL | 6.38s | 4.77s | 1.34× |
| Z-Image | 10.55s | 2.91s | 3.63× |
Warm runs
Models are loaded by an initial generation. Subsequent runs use a new seed for a fresh result.
Comparison
Nox versus unaltered ComfyUI. Time runs from request submission to completion, excluding output download. Speedup = ComfyUI time ÷ Nox time.
Pipeline settings and full results: Source report
Supported image & video models
Image
- Krea2
- Ideogram 4
- Anima
- FLUX.2 Klein 4B & 9B
- SDXL
- Z-Image
- Qwen Image 2512
- Qwen Image Edit 2511
- Qwen Image 2.1
Video
- LTX 2.3
- LTX 2.5
- MiniMax H3





