Changes the exposure from dark to bright, with a range from -6 to +6. As always, results can vary depending on the video and prompt, so keep that in mind when using the slider.

This was trained at rank 1 using only the last two layers, specifically to keep the slider focused on exposure and avoid changing other aspects of the video. That also makes the LoRA ridiculously tiny: only 213 KB.

One thing to keep in mind is that the model itself is very capable and multimodal. If changing the lighting makes something else in the scene logically different, it may sometimes adapt other details accordingly. For example, if an outdoor scene becomes extremely dark, the model might interpret that as nighttime or assume there is some kind of roof or obstruction causing the darkness. Likewise, increasing the exposure can reveal or emphasize details that make sense for the new lighting.

If you explicitly describe what you want in the prompt, the model should follow it. So if you want a specific lighting situation without changing the scene, describe it clearly in the prompt.

In short: the slider controls exposure, but the model may use its understanding of the scene to make lighting-related adjustments that it considers reasonable.