is there anyway to take this jupyter latent blending notebook

[gsgoldma]

and turn it into a script or extension?

https://colab.research.google.com/drive/1I77--5PS6C-sAskl9OggS1zR0HLKdq1M?usp=sharing#scrollTo=jgZQj-tE6GWW

it works really well

!pip install wget
import wget
import os

import requests
model = 'v2-1_768-ema-pruned' #@param ["v2-1_512-ema-pruned", "v2-1_768-ema-pruned", "v1.5"]
#@markdown Optionally, specify your own checkpoint below. Make sure to select the correct model above.
url_ckpt = "" #@param {type:"string"}

if len(url_ckpt) < 1:
if model == "v2-1_512-ema-pruned":
url_ckpt = "https://huggingface.co/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.ckpt"
fp_config = 'latentblending/configs/v2-inference.yaml'
elif model == "v2-1_768-ema-pruned":
url_ckpt = "https://huggingface.co/stabilityai/stable-diffusion-2-1/resolve/main/v2-1_768-ema-pruned.ckpt"
fp_config = 'latentblending/configs/v2-inference-v.yaml'

Check that the supplied URLs exist.

response = requests.head(url_ckpt)
if response.status_code != 200 and response.status_code != 302:
raise ValueError(f"url_ckpt could not be downloaded: {url_ckpt} gives {response.status_code}")
fp_ckpt = 'model.ckpt'
wget.download(url_ckpt, fp_ckpt)
assert os.path.isfile(fp_ckpt), "model download has failed."

if model == "v2-1_512-ema-pruned":
fp_config = 'latentblending/configs/v2-inference.yaml'
elif model == "v2-1_768-ema-pruned":
fp_config = 'latentblending/configs/v2-inference-v.yaml'
elif model == 'v1.5':
fp_config = 'latentblending/configs/v1-inference.yaml'

print(f"url_ckpt: {url_ckpt} fp_config {fp_config}")

installs

!pip install open-clip-torch
!pip install omegaconf
!pip install fastcore -U
!pip install Pillow
!pip install ffmpeg-python
!pip install einops
!pip install gradio

import os, sys
from subprocess import getoutput

Xformers

os.system("pip install --extra-index-url https://download.pytorch.org/whl/cu113 torch torchvision==0.13.1+cu113")
os.system("pip install triton==2.0.0.dev20220701")
gpu_info = getoutput('nvidia-smi')
if("A10G" in gpu_info):
os.system(f"pip install -q https://github.com/camenduru/stable-diffusion-webui-colab/releases/download/0.0.15/xformers-0.0.15.dev0+4c06c79.d20221205-cp38-cp38-linux_x86_64.whl")
elif("T4" in gpu_info):
os.system(f"pip install -q https://github.com/camenduru/stable-diffusion-webui-colab/releases/download/0.0.15/xformers-0.0.15.dev0+1515f77.d20221130-cp38-cp38-linux_x86_64.whl")

!pip install pytorch_lightning
!pip install transformers

Get Latent Blending from git / pull

!git clone https://github.com/lunarring/latentblending
!cd latentblending; git pull; cd ..
sys.path.append("/content/latentblending")

Imports

import torch
import numpy as np
import warnings
warnings.filterwarnings('ignore')
import warnings
import torch
from tqdm.auto import tqdm
from PIL import Image
import torch
from typing import Callable, List, Optional, Union
from latent_blending import LatentBlending, add_frames_linear_interp, get_time, yml_save, LatentBlending, compare_dicts
from stable_diffusion_holder import StableDiffusionHolder
from gradio_ui import BlendingFrontend
import gradio as gr

torch.set_grad_enabled(False)
torch.backends.cudnn.benchmark = False

#%% First let us spawn a stable diffusion holder
device = "cuda"

sdh = StableDiffusionHolder(fp_ckpt, fp_config, device)

from latent_blending import get_time, yml_save, LatentBlending, add_frames_linear_interp, compare_dicts
from gradio_ui import BlendingFrontend

import gradio as gr

if name == "main":

self = BlendingFrontend(sdh) # Yes this is possible in python and yes it is an awesome trick

with gr.Blocks() as demo:
    with gr.Row():
        prompt1 = gr.Textbox(label="prompt 1")
        prompt2 = gr.Textbox(label="prompt 2")
        negative_prompt = gr.Textbox(label="negative prompt")          
        
    with gr.Row():
        nmb_branches_final = gr.Slider(5, 125, self.nmb_branches_final, step=4, label='nmb trans images', interactive=True) 
        height = gr.Slider(256, 2048, self.height, step=128, label='height', interactive=True)
        width = gr.Slider(256, 2048, self.width, step=128, label='width', interactive=True) 
        
    with gr.Row():
        num_inference_steps = gr.Slider(5, 100, self.num_inference_steps, step=1, label='num_inference_steps', interactive=True)
        branch1_influence = gr.Slider(0.0, 1.0, self.branch1_influence, step=0.01, label='branch1_influence', interactive=True) 
        guidance_scale = gr.Slider(1, 25, self.guidance_scale, step=0.1, label='guidance_scale', interactive=True) 

    with gr.Row():
        depth_strength = gr.Slider(0.01, 0.99, self.depth_strength, step=0.01, label='depth_strength', interactive=True) 
        duration = gr.Slider(0.1, 30, self.duration, step=0.1, label='video duration', interactive=True) 
        guidance_scale_mid_damper = gr.Slider(0.01, 2.0, self.guidance_scale_mid_damper, step=0.01, label='guidance_scale_mid_damper', interactive=True) 
        
    with gr.Row():
        seed1 = gr.Number(42, label="seed 1", interactive=True)
        b_newseed1 = gr.Button("randomize seed 1", variant='secondary')
        seed2 = gr.Number(420, label="seed 2", interactive=True)
        b_newseed2 = gr.Button("randomize seed 2", variant='secondary')
    with gr.Row():
        b_compute_transition = gr.Button('compute transition', variant='primary')
    
    with gr.Row():
        img1 = gr.Image(label="1/5")
        img2 = gr.Image(label="2/5")
        img3 = gr.Image(label="3/5")
        img4 = gr.Image(label="4/5")
        img5 = gr.Image(label="5/5")
    
    with gr.Row():
        vid_transition = gr.Video()
    
    # Bind the on-change methods
    depth_strength.change(fn=self.change_depth_strength, inputs=depth_strength)
    num_inference_steps.change(fn=self.change_num_inference_steps, inputs=num_inference_steps)
    nmb_branches_final.change(fn=self.change_nmb_branches_final, inputs=nmb_branches_final)
    
    guidance_scale.change(fn=self.change_guidance_scale, inputs=guidance_scale)
    guidance_scale_mid_damper.change(fn=self.change_guidance_scale_mid_damper, inputs=guidance_scale_mid_damper)
    
    height.change(fn=self.change_height, inputs=height)
    width.change(fn=self.change_width, inputs=width)
    negative_prompt.change(fn=self.change_negative_prompt, inputs=negative_prompt)
    seed1.change(fn=self.change_seed1, inputs=seed1)
    seed2.change(fn=self.change_seed2, inputs=seed2)
    duration.change(fn=self.change_duration, inputs=duration)
    branch1_influence.change(fn=self.change_branch1_influence, inputs=branch1_influence)

    b_newseed1.click(self.randomize_seed1, outputs=seed1)
    b_newseed2.click(self.randomize_seed2, outputs=seed2)
    b_compute_transition.click(self.compute_transition, 
                               inputs=[prompt1, prompt2],
                               outputs=[img1, img2, img3, img4, img5, vid_transition])
    
demo.launch(share=self.share, inbrowser=True, inline=False, debug=True)