A tiny LoRA with a very specific job: controlling how much of the scene is in focus.

It’s only rank 1, around 650 KB, and was trained on just layers 38–43. Small file, surprisingly noticeable effect.

The slider is pretty straightforward:

Negative values → very shallow depth of field
Positive values → deep focus / everything sharp

Around -4 to +4 is where I usually work, but it can respond well beyond that. Values around ±6 can produce stronger effects depending on the model, prompt, and how extreme you want the result.

On the negative side, it can push the image toward a very shallow DOF with strong background and foreground separation, giving you some really nice cinematic bokeh. It also works surprisingly well for still images, not just video.

On the positive side, the focus opens up and more of the scene stays sharp, eventually reaching a much more deep-focus / everything-in-focus look.

The fun part is that it doesn't need a huge LoRA to do it. Just a rank 1, ~650 KB file trained on a small set of layers.

Negative = shallow DOF / strong bokeh
0 = natural focus
Positive = deep DOF / everything sharp

As always, the exact sweet spot depends on the model and prompt, so the values are more of a starting point than a hard rule.