Text-to-Video
Diffusers
TuneAVideoPipeline
tune-a-video
mo-di-bear-guitar / config.yaml
jayw's picture
First model version
1a0adef
pretrained_model_path: /group/30042/zhangjiewu/downloads/mo-di-diffusion
output_dir: ./outputs/disney_lr3e-5_seed33
train_data:
video_path: data/bear-guitar.mp4
prompt: a bear is playing guitar
n_sample_frames: 8
width: 512
height: 512
sample_start_idx: 0
sample_frame_rate: 1
validation_data:
prompts:
- a handsome prince is playing guitar in the summer forest, modern disney style
- a magical princess is playing guitar on the beach, modern disney style
- a rabbit is playing guitar, modern disney style
- a baby is playing guitar, modern disney style
video_length: 8
width: 512
height: 512
num_inference_steps: 50
guidance_scale: 7.5
validation_steps: 100
trainable_modules:
- attn1.to_q
- attn2.to_q
- attn_temp
train_batch_size: 1
max_train_steps: 500
learning_rate: 3.0e-05
scale_lr: false
lr_scheduler: constant
lr_warmup_steps: 0
adam_beta1: 0.9
adam_beta2: 0.999
adam_weight_decay: 0.01
adam_epsilon: 1.0e-08
max_grad_norm: 1.0
gradient_accumulation_steps: 1
gradient_checkpointing: true
checkpointing_steps: 1000
resume_from_checkpoint: null
mixed_precision: fp16
use_8bit_adam: false
enable_xformers_memory_efficient_attention: true
seed: 33