license: apache-2.0
language:
- en
base_model:
- genmo/mochi-1-preview
pipeline_tag: text-to-video
tags:
- jinx
- arcane
- mochi
- diffusion
Fine-Tuning Mochi-Sota Text-to-Video: Jinx Lora Test
This project demonstrates the fine-tuning of the Mochi-Sota Text-to-Video model using a LoRA (Low-Rank Adaptation) approach, focusing on the character Jinx from the League of Legends universe. The goal was to adapt the model to generate dynamic, character-specific video sequences with consistent visual and motion styles.
Training Details
- Model Base: Mochi-Sota Text-to-Video
- Fine-Tuning Dataset: 14 short video clips of Jinx
- Frame Selection: 61 frames extracted from the videos
- Training Hardware: H100 GPU
- Training Duration: 5 hours
This fine-tuning process leverages LoRA to efficiently adapt the model while preserving the core capabilities of the base model.
Results
Below is an example of the generated video output:
Sample Description
Jinx sprints through a dimly lit alley, her vibrant blue hair trailing behind her. She clutches a small, bulging sack tightly against her chest. Dressed in a dark crop top and boots, she moves with chaotic energy, her boots thudding loudly on the pavement. Her mischievous grin flashes briefly as she glances back, her pace never faltering.