Christopher Pietsch cpietsch

🤗 Stable Diffusion for AMD GPUs on Windows using DirectML

Requirements

Convert your WebM to ProRes

Like many of you lucky enough to purchase a PS5, I have been enjoying my time with the new console. One feature I started playing with is the video game capture feature via the "Create" button. I have been capturing my favorite clips here and there and selected the best ones to include a video I want to share. Unfortunately, I ran into trouble. None of my video editing applications (FCPX, DaVinci) support the WebM container. My favorite clips were all captured using WebM.

If you like me have all your favorite clips stuck in WebM then I have a solution for you. FFmpeg! This free software includes a ProRes encoder that simple enough to use works on Davinci Resolve (free version available) and Adobe Premiere! (FCPX of course supports ProRes.) See links below.

MacOS (and Linux)

On the MacOS, you can use the following command (Bash) to convert all WebM files in the current directory to ProRes.

Software

Mainboard firmware: Marlin 2.0.9.2 (fork with config, see Marlin/Configuration{,_adv}.h)
- diff: nvim -d {Marlin,config/Artillery/Genius}/Configuration.h, same for Configuration_adv.h
  - mostly calibration, new extruder and noise-free PWM for parts fan
  - LIN_ADVANCE instead of S_CURVE_ACCELERATION (don't work together)
- before 13.11.2020: Marlin 2.0.5.3 (3dprintbeginner) (branch)
TFT firmware: artillery_tft_fw_1.27.x_patch_9.2 (digant@thingiverse) (TFT-config.ini)
- Put in Marlin mode (long press, persists on reboot) when printing with OctoPrint because TFT shares serial connection.

Kiosk Mode macOS

Create two user accounts
- Administration account
- Kiosk Account - This account will be locked down with Parental Controls
Set up remote management
- We are using Screens 4
- See Screens documentation Configuring macOS for Remote Access
Install kiosk software

We use a role-based Apple ID (ex: webmaster@[institution].org) to manage macOS software. Managing Apps this way saves lots of headaches.

Making a big image zoomable

When you have a giant image and you want to make it easy to pan and zoom without downloading the whole 50MB image into someone's browser, a nice workaround is to cut that image into tiles at different zoom levels and view it as it were a map. An example where I've used this technique is The "Snowpiercer" Scenario.

One way to cut your big image into the requisite tiles is with gdal2tiles.py.

Alternatively, this Node script will do the cutting after you install node-canvas and mkdirp:

const fs = require("fs"),

	[include mainsail.cfg]

	[gcode_macro _CLIENT_VARIABLE]
	variable_use_custom_pos : False ; use custom park coordinates for x,y [True/False]
	variable_custom_park_x : 0.0 ; custom x position; value must be within your defined min and max of X
	variable_custom_park_y : 0.0 ; custom y position; value must be within your defined min and max of Y
	variable_custom_park_dz : 2.0 ; custom dz value; the value in mm to lift the nozzle when move to park position
	variable_retract : 1.0 ; the value to retract while PAUSE
	variable_cancel_retract : 5.0 ; the value to retract while CANCEL_PRINT
	variable_speed_retract : 35.0 ; retract speed in mm/s

	import torch

	#USE AT YOUR OWN RISK

	#local path to runwayML SD 1.5 checkpoint (https://huggingface.co/runwayml/stable-diffusion-v1-5)
	ckpt_15 = "./v1-5-pruned-emaonly.ckpt"

	#local path to StabilityAI finetuned autoencoder (https://huggingface.co/stabilityai/sd-vae-ft-mse)
	ckpt_vae = "./vae-ft-mse-840000-ema-pruned.ckpt"

	import copy
	import os
	import click
	import numpy as np
	import PIL.Image
	import PIL.ImageOps
	import torch
	import dnnlib
	import legacy
	from tqdm import tqdm

	#!/bin/bash

	if [ -z "$2" ];then
	echo 'USAGE:

	denoise input.mov output.mov

	OR

	denoise input.mov output.mov [ambient-noise-start-time] [ambient-noise-duration] [sox-noisered-amount] [sox-norm-param]

	.Rhistory
	.Rproj.user
	covid19-berlin.Rproj
	.Renviron