|
num_frames = 16 |
|
fps = 24 // 3 |
|
image_size = (256, 256) |
|
|
|
|
|
model = dict( |
|
type="STDiT-XL/2", |
|
space_scale=0.5, |
|
time_scale=1.0, |
|
enable_flash_attn=True, |
|
enable_layernorm_kernel=True, |
|
from_pretrained="PRETRAINED_MODEL", |
|
) |
|
vae = dict( |
|
type="VideoAutoencoderKL", |
|
from_pretrained="stabilityai/sd-vae-ft-ema", |
|
micro_batch_size=4, |
|
) |
|
text_encoder = dict( |
|
type="t5", |
|
from_pretrained="DeepFloyd/t5-v1_1-xxl", |
|
model_max_length=120, |
|
) |
|
scheduler = dict( |
|
type="iddpm", |
|
num_sampling_steps=100, |
|
cfg_scale=7.0, |
|
cfg_channel=3, |
|
) |
|
dtype = "bf16" |
|
|
|
|
|
prompt_path = "./assets/texts/t2v_samples.txt" |
|
prompt = None |
|
|
|
|
|
batch_size = 1 |
|
seed = 42 |
|
save_dir = "./samples/samples/" |
|
|