Squish LoRA for Wan2.1
| Property | Value |
|---|---|
| Base Model | Wan2 .1 14B I2V 480p |
| Training Data | 1.5 minutes of video (20 clips) |
| Training Epochs | 18 |
| Model URL | https://huggingface.co/Remade-AI/Squish |
What is Squish?
Squish is a specialized LoRA model trained on the Wan2.1 14B I2V 480p base model that enables the creation of videos showing objects being squished. This innovative model can transform static images into dynamic videos featuring a satisfying squishing effect, applicable to a wide range of subjects including animals, vehicles, and people.
Implementation Details
The model implements a specific workflow using the Diffusers library and requires particular prompt engineering for optimal results. It operates at 480p resolution and supports 81-frame video generation with customizable guidance scaling and inference steps.
- Integrated with ComfyUI through modified Kijai's Wan Video Wrapper workflow
- Supports various resolution outputs with automatic aspect ratio calculation
- Implements CPU offload support for low-VRAM environments
Core Capabilities
- Generates squishing effect animations from static images
- Supports multiple object types with consistent results
- Processes 81 frames at 16 FPS
- Includes automatic resolution scaling
- Features trigger phrase "sq41sh squish effect"
Frequently Asked Questions
Q: What makes this model unique?
The model specializes in creating realistic squishing animations using a specific prompt structure and trigger words, trained specifically on squishing motion data.
Q: What are the recommended use cases?
The model is ideal for creating entertaining animations of objects being squished, particularly useful for creative content creation, entertainment purposes, and demonstrating physical deformation effects.
