A long exposure and motion blur lora. Training data is shot by myself.

For a better context just naming what it's trained on. Everything was shot at 5-10 seconds exposure.

1. Typical tri-pod shots of moving cars or S-Bahn trains in the dark for linear light streaks.

2. Non locked camera. Walking around or driving a bike at night in the city while capturing.

3. Waving around LED lights or mobile phones in random patterns in front of the camera.

4. Children's firecrackers that jump around on the ground emitting light.

5. A ghetto water rig where water flows down through a hose and the camera mounted to it. I flipped it by 90 degrees to go from left to right instead of down to bottom though.

6. Same rig as above but at late afternoon where it was still bright enough. So far I found motion blur hard to do in Flux and this way it somehow translates well. (The guy on the bike example shows that it kinda understands it without directly prompting for it.)

Edit: for some reason this one seems to work with Euler the best. No idea why.