|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
pipeline_tag: text-to-video |
|
|
tags: |
|
|
- art |
|
|
- text-generation-inference |
|
|
--- |
|
|
introduction:(this is a motion lora model/Research on motion models about Beam and shuttle sence) |
|
|
1.As the Apache license2.0 license |
|
|
2.any commercial legal disputes arising from the use of this model shall be borne by yourself. |
|
|
3.if you install this model that implies that you accept the terms. |
|
|
--- |
|
|
*2example with 2K result has been released on the youtube(https://www.youtube.com/watch?v=s_NdPdkbNqU)twitter(https://twitter.com/Drex1165890669/status/1758196689968480430) |
|
|
|
|
|
**motion lora model configuration |
|
|
|
|
|
1.train in 1024x576size 256rank 16framesx5 1500steps |
|
|
|
|
|
2.six videos input with folder training with 3 types of text config |
|
|
|
|
|
3.trian prompt:"a highly quality clear best realistic video of high speed shuttle with light" |
|
|
|
|
|
4.training model:epicrealism_naturalSinRC1VAE.safetensors, inference model:epicrealism_naturalSinRC1VAE.safetensors |
|
|
|
|
|
5.additionally necessary specify parameters：use_offset_noise: true |
|
|
|
|
|
6.learning_rate/learning_rate_spatial/adam_weight_decay:between 5x-20xoriginal learning_rate=depending on your training set counts |
|
|
|
|
|
7.Suitable for Animatediff workflow, weight recommended in between 0.5-1 |
