The video your friend shot on their phone, as a style LoRA for MiniMax H3.

A frame wider than anyone would compose, flat phone light, the subject pushed to the edge, a hand that shakes while your friend laughs.

How to use

Trigger: celsint

Weight 0.65 when faces are close: they hold and the phone look stays. Go up to 0.8 when people are far away, turned around or not in the shot: the look gets stronger, but close faces start to drift.

Put the style in the first sentence of the H3 prompt: [Shot 1] Live-action in the celsint style, a medium shot at chest height...

Don't describe the capture (phone, handheld, shake, exposure, grain). The trigger carries it.

Describe the scene: who it is, where they stand in the frame, what makes them react, and what the friend filming is doing. A friend who runs behind, gets pulled by the wrist or bumped by a guest shakes the frame the way a real one does.

Best in close scenes with people: a house party, a street at night, a supermarket aisle. Run it alone, without other anatomy or action LoRAs.

Tested on MiniMax H3 FL2VA pruned int8 ConvRot in ComfyUI, text-to-video: 20 steps, res_multistep, simple, 768x1344, 5 seconds. Trained on FL2VA pruned.

Original style. Use what you make with it however you like, commercially included, and credit the style. Reuploading this LoRA, or training another model on its outputs, needs our permission first: send a message.

Commissions are open: bring me your look or your camera and I'll cut an H3 LoRA to measure. Send me a message.