This LoRA allow you to squish objects, not only when brought to the camera, but at a distance too! Unlike vanilla squish, you can use a single hand alone without much prompting!
Prompt:
In the video, an [object] is presented. A person's long right arm appears from the side, stretches all way to the [object] and grabs the [object]. The [object] is held in the person’s right hand in the distance. The person then presses on the [object], causing a sq41sh squish effect. The person keeps pressing down on the [object], further showing the sq41sh squish effect.
My usage settings are straightforward: 3.5 CFG, shift 8.0 (default), dpm++-sde scheduler, 25 (13+12) steps. No speedup, or otherwise additional LoRAs have been used. Made with Kijai's wrapper.
I used uncond skip of the 10th block (SLG), to enhance the video quality a bit. No other enhancements have been applied for the gallery generations. It all works greatly on 512x512, 640p and 720p.
For details, see the workflows in the videos metadata or their duplicates on GitHub.
Now TL;DR. Below I present my struggles with this LoRA creation. I think, it may be useful for the fellow LoRA trainers and techy users in general.
This LoRA allow you to squish objects, not only when brought to the camera, but at a distance too! Unlike vanilla squish, you can use a single hand alone without much prompting!
Prompt:
In the video, an [object] is presented. A person's long right arm appears from the side, stretches all way to the [object] and grabs the [object]. The [object] is held in the person’s right hand in the distance. The person then presses on the [object], causing a sq41sh squish effect. The person keeps pressing down on the [object], further showing the sq41sh squish effect.
My usage settings are straightforward: 3.5 CFG, shift 8.0 (default), dpm++-sde scheduler, 25 (13+12) steps. No speedup, or otherwise additional LoRAs have been used. Made with Kijai's wrapper.
I used uncond skip of the 10th block (SLG), to enhance the video quality a bit. No other enhancements have been applied for the gallery generations. It all works greatly on 512x512, 640p and 720p.
For details, see the workflows in the videos metadata or their duplicates on GitHub.
Now TL;DR. Below I present my struggles with this LoRA creation. I think, it may be useful for the fellow LoRA trainers and techy users in general.