CogVideoX1.0-LoRA-Arcane-v1
| Property | Value |
|---|---|
| Base Model | THUDM/CogVideoX-5b |
| Training Steps | 17,000 |
| Dataset Size | 136 videos (49x720x480) |
| Papers | CogVideoX & CogVideo Research Papers |
| License | Research/Non-commercial Use Only |
What is CogVideoX1.0-LoRA-Arcane-v1?
CogVideoX1.0-LoRA-Arcane-v1 is a specialized LoRA adaptation of the CogVideoX model, fine-tuned to generate videos in the style of the Arcane animated series. This model represents a fan project developed for research purposes, enabling text-to-video generation with specific aesthetic qualities reminiscent of the show.
Implementation Details
The model was trained using adamw optimizer with a learning rate of 1e-4 and cosine_with_restarts scheduler. It features a rank/alpha of 128/128 and was trained on a curated dataset of 136 videos. The implementation requires the base CogVideoX-5b model and supports specific character tokens for enhanced generation control.
- Compatible exclusively with CogVideoX 1.0
- Includes specialized character tokens for consistent character generation
- Trained with batch size 1 for optimal results
- No quantization or safe attention recommended for best quality
Core Capabilities
- Text-to-video generation in Arcane-inspired style
- Character-specific generation using dedicated tokens
- Expression and lighting control through prompt engineering
- Smooth transition animations between frames
Frequently Asked Questions
Q: What makes this model unique?
This model specifically targets the Arcane animation style and includes character-specific tokens for more controlled generation, though with the caveat that character reproduction accuracy is limited.
Q: What are the recommended use cases?
The model is intended for research and non-commercial purposes only, specifically for generating Arcane-style video content. It's particularly effective for creating mood-driven character animations with specific lighting and expression transitions.
