takuma104

#!/bin/bash

models=("canny" "depth" "pose" )

for model in "${models[@]}"
do
    echo $model
    python merge_controlnet_diff.py --sd15 ../wd-1-5-beta2/checkpoints/wd-1-5-beta2-fp32.safetensors --control diff_control_wd15beta2_$model.safetensors --dst control_wd15beta2_$model.safetensors --fp16
done

takuma104

import torch
import torch.nn.functional as F
import safetensors.torch
import sys


def load_checkpoint(fn):
    if fn.endswith(".safetensors"):
        checkpoint = safetensors.torch.load_file(fn)
    else:

takuma104

import torch
import pytest
import PIL.Image
import numpy as np
from diffusers.utils import (
    PIL_INTERPOLATION,
)

# `batch` = batch_size * num_images_per_prompt
# When the input is a single image or a tensor with b==1, repeat it the `batch` times.

takuma104

import torch
from diffusers import UNet2DConditionModel

mem_bytes = torch.cuda.max_memory_allocated()
torch.cuda.reset_peak_memory_stats()
controlnet = UNet2DConditionModel.from_pretrained("takuma104/control_sd15_canny", subfolder="controlnet").cuda()
mem_bytes = torch.cuda.max_memory_allocated()
print(mem_bytes)

takuma104

takuma104

# Diffusers' ControlNet Implementation Subjective Evaluation
# https://github.com/takuma104/diffusers/tree/controlnet

import einops
import numpy as np
import pytest
import torch

from diffusers import StableDiffusionControlNetPipeline

takuma104

# from https://github.com/lllyasviel/ControlNet/blob/main/gradio_canny2image.py

from share import *
import config

import cv2
import einops
import gradio as gr
import numpy as np
import torch

takuma104

Reproducible generation with xFormers for SD2.x or variant models

More details and background: Generating (almost) reproducible pictures using Diffusers with xFormers

Limitation

Unfortunately, Flash Attention won't accept SD1.x attention shapes, but will accept SD2.x or variants.

Test Result

takuma104

import torch
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler, EulerAncestralDiscreteScheduler
import matplotlib.pyplot as plt
from PIL import Image
import numpy as np
import xformers
import xformers.ops

plt.rcParams["figure.figsize"] = (10,5)
plt.rcParams['figure.facecolor'] = 'white'

takuma104

import torch
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
import matplotlib.pyplot as plt
from PIL import Image
import numpy as np

plt.rcParams["figure.figsize"] = (10,5)
plt.rcParams['figure.facecolor'] = 'white'

def generate_tuxedo_cat_picture(fn_prefix, seed=0):