htoyryla’s gists

htoyryla / sd1mp3.mpy

Created October 10, 2022 15:29

Stable diffusion text2image assigning combined embeddings to each UNet block

	# stable diffusion tool

	# @htoyryla github twitter instagram

	# requires diffusers 0.3.0 and a trained model

	# relies heavily on code from https://github.com/huggingface/diffusers

	# neurokuvatreenit, stable diffusion, example 1b using LDM scheduler and 1c saving image at each iteration
	# 1d: multiple subprompts with weights

htoyryla / sd1mp2.py

Last active October 10, 2022 08:16

Stable diffusion text2image assigning different prompts to each UNet block EXPERIMENTAL

	# stable diffusion tool

	# @htoyryla github twitter instagram

	# requires diffusers 0.4.0 and a trained model

	# relies heavily on code from https://github.com/huggingface/diffusers

	# neurokuvatreenit, stable diffusion, example 1b using LDM scheduler and 1c saving image at each iteration
	# 1d: multiple subprompts with weights

htoyryla / stadit.py

Last active September 21, 2022 07:18

Stable diffusion tool, image synthesis, img2img and prompt weights

	# stable diffusion tool

	# @htoyryla github twitter instagram

	# requires diffusers 0.3.0 and a trained model

	# relies heavily on code from https://github.com/huggingface/diffusers

	# for prompt weighting use prompts like "a subprompt like this:10 / in the style of that:25"

htoyryla / i2it.py

Last active September 8, 2022 15:07

Script for image2image using stablediffusion using incremental strength

	from torch import autocast
	import torch
	import requests
	from PIL import Image
	import sys
	from diffusers import StableDiffusionImg2ImgPipeline #, preprocess
	import random
	import argparse

	# image to image generation using StableDiffusion 0.3.0 dev

htoyryla / i2is.py

Last active August 27, 2022 16:06

script to generate 100 frames transition from init image towards the promp with Stablediffusion

	from torch import autocast
	import torch
	import requests
	from PIL import Image
	import sys
	from image_to_image import StableDiffusionImg2ImgPipeline, preprocess
	import random

	# script to generate 100 frames transition from init image towards the prompt

htoyryla / vid2.py

Created August 26, 2022 11:49

Simple script to read frames from a video and use as init images in stablediffusion

	#
	# Usage:
	#
	# python vid2.py "prompt" video_in_path output_dir [strength scale seed]
	#
	# requires image_to_image.py from https://github.com/huggingface/diffusers/blob/main/examples/inference/image_to_image.py
	#


	from torch import autocast

htoyryla / mdisco3.py

Created August 6, 2022 08:21

Basic disco diffusion with masks, command line only, config read from ini file

	# %%
	# !! {"metadata":{
	# !! "id": "view-in-github",
	# !! "colab_type": "text"
	# !! }}
	"""
	<a href="https://colab.research.google.com/github/alembics/disco-diffusion/blob/main/Disco_Diffusion.ipynb" target="_parent"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a>
	"""

	# %%

htoyryla / _ht_gan.txt

Last active June 11, 2018 11:13

HT-GAN including AE and growing image size

	code for a gan trainer
	adding an encoder to assist in generator training
	htoyryla 8 Jun 2018

	support for progressive training with larger image size
	htoyryla 11.6.2018

htoyryla / neural-equalchannels.lua

Created February 14, 2018 15:19

	-- modified version of @jcjohnson's neural-style
	-- by @htoyryla 14 Feb 2018
	-- allows exploring the effect of equalizing the activations of different channels in a style layer


	require 'torch'
	require 'nn'
	require 'image'
	require 'optim'

htoyryla / neural_bestchannels.lua

Last active March 27, 2018 09:58

	-- modified version of @jcjohnson's neural-style
	-- by @htoyryla 13 Feb 2018
	-- allows giving emphasis to nc best channel(s) in each style layer
	-- use -style_layers to select layer as usual, using a single layer is recommended
	-- -nc to set how many of the best channels are used per layer
	-- during target capture, tests the model using the style image
	-- and selects nc channels with strongest activations to be given emphasis during iterations
	-- not tested with multiple style images

	require 'torch'

Hannu Töyrylä htoyryla