# Copyright 2022 Lunar Ring. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import os, sys import torch torch.backends.cudnn.benchmark = False 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 movie_util import MovieSaver from typing import Callable, List, Optional, Union from latent_blending import get_time, yml_save, LatentBlending, add_frames_linear_interp from stable_diffusion_holder import StableDiffusionHolder torch.set_grad_enabled(False) import gradio as gr import copy """ experiment with slider as output -> does it change in the browser? guidance scale has no effect get a movie as result seed bug: also shows changes from before mid compression scaler can destroy tree """ #%% def compare_dicts(a, b): """ Compares two dictionaries a and b and returns a dictionary c, with all keys,values that have shared keys in a and b but same values in a and b. The values of a and b are stacked together in the output. Example: a = {}; a['bobo'] = 4 b = {}; b['bobo'] = 5 c = dict_compare(a,b) c = {"bobo",[4,5]} """ c = {} for key in a.keys(): if key in b.keys(): val_a = a[key] val_b = b[key] if val_a != val_b: c[key] = [val_a, val_b] return c class BlendingFrontend(): def __init__(self): self.use_debug = False self.share = True self.height = 512 self.width = 512 self.num_inference_steps = 30 self.depth_strength = 0.25 self.seed1 = 42 self.seed2 = 420 self.guidance_scale = 4.0 self.guidance_scale_mid_damper = 0.5 self.mid_compression_scaler = 1.2 self.prompt1 = '' self.prompt2 = '' self.dp_base = '/home/lugo/latentblending' self.list_settings = [] self.state_prev = {} self.state_current = {} self.showing_current = True self.imgs_show_last = [] self.imgs_show_current = [] if not self.use_debug: self.init_diffusion() def init_diffusion(self): fp_ckpt = "../stable_diffusion_models/ckpt/v2-1_512-ema-pruned.ckpt" fp_config = 'configs/v2-inference.yaml' # fp_ckpt = "../stable_diffusion_models/ckpt/v2-1_768-ema-pruned.ckpt" # fp_config = 'configs/v2-inference-v.yaml' sdh = StableDiffusionHolder(fp_ckpt, fp_config, height=self.height , width=self.width, num_inference_steps=self.num_inference_steps) self.lb = LatentBlending(sdh) self.use_debug = False def change_depth_strength(self, value): self.depth_strength = value print(f"changed depth_strength to {value}") def change_num_inference_steps(self, value): self.num_inference_steps = value print(f"changed num_inference_steps to {value}") def change_guidance_scale(self, value): self.guidance_scale = value self.lb.set_guidance_scale(value) print(f"changed guidance_scale to {value}") def change_guidance_scale_mid_damper(self, value): self.guidance_scale_mid_damper = value print(f"changed guidance_scale_mid_damper to {value}") def change_mid_compression_scaler(self, value): self.mid_compression_scaler = value print(f"changed mid_compression_scaler to {value}") def change_height(self, value): self.height = value print(f"changed height to {value}") def change_width(self, value): self.width = value print(f"changed width to {value}") def change_prompt1(self, value): self.prompt1 = value # print(f"changed prompt1 to {value}") def change_prompt2(self, value): self.prompt2 = value # print(f"changed prompt2 to {value}") def change_seed1(self, value): self.seed1 = int(value) def change_seed2(self, value): self.seed2 = int(value) def randomize_seed1(self): seed = np.random.randint(0, 10000000) self.change_seed1(seed) print(f"randomize_seed1: new seed = {self.seed1}") return seed def randomize_seed2(self): seed = np.random.randint(0, 10000000) self.change_seed2(seed) print(f"randomize_seed2: new seed = {self.seed2}") return seed def run(self, x): print("STARTING DIFFUSION!") self.state_prev = self.state_current.copy() self.state_current = self.get_state_dict() # Copy last iteration self.imgs_show_last = copy.deepcopy(self.imgs_show_current) if self.use_debug: list_imgs = [(255*np.random.rand(200,200,3)).astype(np.uint8) for l in range(5)] self.imgs_show_current = copy.deepcopy(list_imgs) return list_imgs # FIXME TODO ASSERTS self.lb.sdh.height = self.height self.lb.sdh.width = self.width # list_nmb_branches = [2, 6, 15] # list_injection_strength = [0.0, self.depth_strength, 0.9] # self.lb.setup_branching( # num_inference_steps = self.num_inference_steps, # list_nmb_branches = list_nmb_branches, # list_injection_strength = list_injection_strength # ) self.lb.autosetup_branching( depth_strength = self.depth_strength, num_inference_steps = self.num_inference_steps, nmb_branches_final = 13, nmb_mindist = 2) self.lb.set_prompt1(self.prompt1) self.lb.set_prompt2(self.prompt2) self.lb.guidance_scale = self.guidance_scale self.lb.guidance_scale_mid_damper = self.guidance_scale_mid_damper self.lb.mid_compression_scaler = self.mid_compression_scaler fixed_seeds = [self.seed1, self.seed2] imgs_transition = self.lb.run_transition(fixed_seeds=fixed_seeds) imgs_transition = [Image.fromarray(l) for l in imgs_transition] print(f"DONE DIFFUSION! Resulted in {len(imgs_transition)} images") nmb_imgs_show = 5 idx_list = np.arange(0, nmb_imgs_show).astype(np.int32)*3 list_imgs = [] for j in idx_list: list_imgs.append(imgs_transition[j]) self.imgs_show_current = copy.deepcopy(list_imgs) return list_imgs def save(self): if self.lb.tree_final_imgs[0] is None: return print("save is called!") dp_img = os.path.join(self.dp_base, get_time("second")) imgs_transition = self.lb.tree_final_imgs self.lb.write_imgs_transition(dp_img, imgs_transition) fps = 20 # Let's get more cheap frames via linear interpolation (duration_transition*fps frames) imgs_transition_ext = add_frames_linear_interp(imgs_transition, 5, fps) # Save as MP4 fp_movie = os.path.join(dp_img, "movie_lowres.mp4") if os.path.isfile(fp_movie): os.remove(fp_movie) ms = MovieSaver(fp_movie, fps=fps) for img in tqdm(imgs_transition_ext): ms.write_frame(img) ms.finalize() return fp_movie def get_state_dict(self): state_dict = {} grab_vars = ['prompt1', 'prompt2', 'seed1', 'seed2', 'height', 'width', 'num_inference_steps', 'depth_strength', 'guidance_scale', 'guidance_scale_mid_damper', 'mid_compression_scaler'] for v in grab_vars: state_dict[v] = getattr(self, v) return state_dict def compare_last(self): if len(self.state_prev) == 0 or len(self.state_current) == 0: return "" if self.showing_current: # inject the last images that were shown and return str of changes str_fill = "showing last version: " list_return = self.imgs_show_last idx = 0 verb = 'was' self.showing_current = False elif not self.showing_current: # inject the current images and show no string str_fill = "showing current version: " verb = 'is' idx = 1 list_return = self.imgs_show_current self.showing_current = True dict_diff = compare_dicts(self.state_prev, self.state_current) for key in dict_diff: str_fill += f"{key} {verb} {dict_diff[key][idx]}, " str_fill = str_fill[:-2] list_return.extend([str_fill]) return list_return self = BlendingFrontend() with gr.Blocks() as demo: with gr.Row(): text1 = gr.Textbox(label="prompt 1") text2 = gr.Textbox(label="prompt 2") with gr.Row(): depth_strength = gr.Slider(0.01, 0.99, self.depth_strength, step=0.01, label='depth_strength', interactive=True) guidance_scale = gr.Slider(1, 25, self.guidance_scale, step=0.1, label='guidance_scale', 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) mid_compression_scaler = gr.Slider(1.0, 2.0, self.mid_compression_scaler, step=0.01, label='mid_compression_scaler', interactive=True) with gr.Row(): num_inference_steps = gr.Slider(5, 100, self.num_inference_steps, step=1, label='num_inference_steps', 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(): b_newseed1 = gr.Button("rand seed 1") seed1 = gr.Number(42, label="seed 1", interactive=True) b_newseed2 = gr.Button("rand seed 2") seed2 = gr.Number(420, label="seed 2", interactive=True) b_compare = gr.Button("compare") b_save = gr.Button('save!') with gr.Row(): b_run = gr.Button('run preview!') 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(): compare_text = gr.Textbox(label="") with gr.Row(): vid = 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) 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) mid_compression_scaler.change(fn=self.change_mid_compression_scaler, inputs=mid_compression_scaler) height.change(fn=self.change_height, inputs=height) width.change(fn=self.change_width, inputs=width) text1.change(fn=self.change_prompt1, inputs=text1) text2.change(fn=self.change_prompt2, inputs=text2) seed1.change(fn=self.change_seed1, inputs=seed1) seed2.change(fn=self.change_seed2, inputs=seed2) b_newseed1.click(self.randomize_seed1, outputs=seed1) b_newseed2.click(self.randomize_seed2, outputs=seed2) b_compare.click(self.compare_last, outputs=[img1, img2, img3, img4, img5, compare_text]) b_run.click(self.run, outputs=[img1, img2, img3, img4, img5]) b_save.click(self.save, outputs=vid) demo.launch(share=self.share)