The code should only requires diffusers==0.18.2.My enviroment is same as https://github.com/haohantianchen/Animate.
p2p
is the directory for Prompt2Prompt codes.
src
is the directory for AnimateDiff codes.
xt_unc
is the directory to save some tensors.
tools
is the directory to do some preprocess.
Install AnimateDiff-PromptTravel follow https://github.com/haohantianchen/Animate.
Then in the root dir, replace the files in src
dir, copy the p2p
dir into the root dir .
bash infer.sh
The default config will generate 16 frames video , loading bg latents from xt_unc/bg_latents.pt
and fg latents from xt_unc/rec_girl_fg_latents.pt
, then blend bg latents with weights w1=[0.9,0.1]
and warp fg latents.
The bg and fg latents are generated by test_nti.py
. fg latents are reconstructed from ref_image.png
using null-text inversion. More details can be seen in readme_image.md
.
More args can be seen in cli.py, generate() function
.