technicalkiwi
/
latentblending
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

new branching setup

This commit is contained in:
Johannes Stelzer 2023-01-08 10:33:11 +01:00
parent cd45b2e585
commit 8bc76b6e3a
3 changed files with 12 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
example1_standard.py
View File

@ -40,18 +40,19 @@ sdh = StableDiffusionHolder(fp_ckpt, fp_config, device)
#%% Next let's set up all parameters #%% Next let's set up all parameters
quality = 'medium' quality = 'medium'
depth_strength = 0.65 # Specifies how deep (in terms of diffusion iterations the first branching happens) depth_strength = 0.35 # Specifies how deep (in terms of diffusion iterations the first branching happens)
fixed_seeds = [69731932, 504430820] fixed_seeds = [69731932, 504430820]
prompt1 = "photo of a beautiful cherry forest covered in white flowers, ambient light, very detailed, magic" # prompt1 = "A person in an open filed of grass watching a television, red colors dominate the scene, eerie light, dark clouds on the horizon, artistically rendered by Richter"
prompt2 = "photo of an golden statue with a funny hat, surrounded by ferns and vines, grainy analog photograph, mystical ambience, incredible detail" prompt1 = "A person in a bar, people around him, a glass of baer, artistically rendered in the style of Hopper"
prompt2 = "A person with a sad expression, looking at a painting of an older man, all in the style of Lucien Freud"
duration_transition = 12 # In seconds duration_transition = 12 # In seconds
fps = 30 fps = 30
# Spawn latent blending # Spawn latent blending
lb = LatentBlending(sdh) lb = LatentBlending(sdh)
lb.autosetup_branching(quality=quality, depth_strength=depth_strength) lb.load_branching_profile(quality=quality, depth_strength=depth_strength)
lb.set_prompt1(prompt1) lb.set_prompt1(prompt1)
lb.set_prompt2(prompt2) lb.set_prompt2(prompt2)

7
example2_inpaint.py
View File

@ -25,7 +25,6 @@ import torch
from tqdm.auto import tqdm from tqdm.auto import tqdm
from diffusers import StableDiffusionInpaintPipeline from diffusers import StableDiffusionInpaintPipeline
from PIL import Image from PIL import Image
import matplotlib.pyplot as plt
import torch import torch
from movie_util import MovieSaver from movie_util import MovieSaver
from typing import Callable, List, Optional, Union from typing import Callable, List, Optional, Union
@ -42,21 +41,21 @@ sdh = StableDiffusionHolder(fp_ckpt, fp_config, device)
#%% Let's first make a source image and mask. #%% Let's first make a source image and mask.
quality = 'medium' quality = 'medium'
deepth_strength = 0.65 #Specifies how deep (in terms of diffusion iterations the first branching happens) depth_strength = 0.65 #Specifies how deep (in terms of diffusion iterations the first branching happens)
duration_transition = 7 # In seconds duration_transition = 7 # In seconds
fps = 30 fps = 30
seed0 = 190791709 seed0 = 190791709
# Spawn latent blending # Spawn latent blending
lb = LatentBlending(sdh) lb = LatentBlending(sdh)
lb.autosetup_branching(quality=quality, deepth_strength=deepth_strength) lb.load_branching_profile(quality=quality, depth_strength=depth_strength)
prompt1 = "photo of a futuristic alien temple in a desert, mystic, glowing, organic, intricate, sci-fi movie, mesmerizing, scary" prompt1 = "photo of a futuristic alien temple in a desert, mystic, glowing, organic, intricate, sci-fi movie, mesmerizing, scary"
lb.set_prompt1(prompt1) lb.set_prompt1(prompt1)
lb.init_inpainting(init_empty=True) lb.init_inpainting(init_empty=True)
lb.set_seed(seed0) lb.set_seed(seed0)
# Run diffusion # Run diffusion
list_latents = lb.run_diffusion(lb.text_embedding1) list_latents = lb.run_diffusion([lb.text_embedding1])
image_source = lb.sdh.latent2image(list_latents[-1]) image_source = lb.sdh.latent2image(list_latents[-1])
mask_image = 255*np.ones([512,512], dtype=np.uint8) mask_image = 255*np.ones([512,512], dtype=np.uint8)

11
example3_multitrans.py
View File

@ -22,7 +22,6 @@ import warnings
import torch import torch
from tqdm.auto import tqdm from tqdm.auto import tqdm
from PIL import Image from PIL import Image
import matplotlib.pyplot as plt
import torch import torch
from movie_util import MovieSaver from movie_util import MovieSaver
from typing import Callable, List, Optional, Union from typing import Callable, List, Optional, Union
@ -32,8 +31,8 @@ torch.set_grad_enabled(False)
#%% First let us spawn a stable diffusion holder #%% First let us spawn a stable diffusion holder
device = "cuda" device = "cuda"
fp_ckpt = "../stable_diffusion_models/ckpt/768-v-ema.ckpt" fp_ckpt = "../stable_diffusion_models/ckpt/v2-1_768-ema-pruned.ckpt"
fp_config = '../stablediffusion/configs/stable-diffusion/v2-inference-v.yaml' fp_config = 'configs/v2-inference-v.yaml'
sdh = StableDiffusionHolder(fp_ckpt, fp_config, device) sdh = StableDiffusionHolder(fp_ckpt, fp_config, device)
@ -56,16 +55,14 @@ list_prompts.append("statue of an ancient cybernetic messenger annoucing good ne
list_seeds = [954375479, 332539350, 956051013, 408831845, 250009012, 675588737] list_seeds = [954375479, 332539350, 956051013, 408831845, 250009012, 675588737]
lb = LatentBlending(sdh) lb = LatentBlending(sdh)
lb.autosetup_branching(quality=quality, depth_strength=depth_strength) lb.load_branching_profile(quality=quality, depth_strength=depth_strength)
fp_movie = "movie_example3.mp4" fp_movie = "movie_example3.mp4"
ms = MovieSaver(fp_movie, fps=fps)
lb.run_multi_transition( lb.run_multi_transition(
fp_movie,
list_prompts, list_prompts,
list_seeds, list_seeds,
ms=ms,
fps=fps, fps=fps,
duration_single_trans=duration_single_trans duration_single_trans=duration_single_trans
) )