Create README.md

README.md

@@ -0,0 +1,52 @@
+A dataset of AI-generated images, selected by the community, for fine tuning Stable Diffusion 1.5 and/or Stable Diffusion 2.0 on particular desired artstyles, 
+WITHOUT using any images owned by major media companies, nor images by artists who are uncomfortable with their works being used to train AI.
+
+What to submit: Your own AI-generated Images paired with a text description of their subject matter.  Filename can be whatever, but both should have the same filename.
+     If you're a conventional digital artist and want to contribute, include solid documentation of your permission for use in the text file.  (see rules below)
+
+Format: Images as .jpeg or .png: 512x512 or preferably 768x768
+        Text files as .txt: with a short list of one-word descriptions about what's in the image. (Not the prompts, describe it in your own words)
+        
+Where to submit: Select the directory most descriptive of the "style" you're going for.  look at a few already-submitted images to make sure it fits. 
+     If youre piece is a different style, something new, prepare at least 4 similarly styled images, and a creative style name, and push a new directory too.
+
+What is going to be done with the images: Once the technical issues get ironed out I will use Hugging Face Diffuser's Dreambooth example script to train the 
+     Stable Diffusion 2.0 model on the 768px images once we have accumulated at least 50 images in a particular style.  I will document the setup, process, and make the 
+     resultant model available here for download.  Once (if) we accumulate on the order of 1000 images, I will begin work on a natively fine-tuned model, incorporating the
+     provided text-image pairs.  
+     Until that point, all images and text descriptions will be available here.  If you wish to use the datasets for your own training projects at any point, the data is 
+     available, and for the purposes of AI training it is free to use.  (Subject to license and Hugging Face T.O.S.)
+     I will not be taking submissions of trained models or other code or utilities.
+
+Basic Rules:
+The first rule of GAI: No images from human artists without the artist's explicit permission. 
+     -In order to document the artist's permission, include a text file with a link to a public statement (like a post on social media for instance) 
+          granting permission for use of that image, or a set of images including the image in question.  
+     -The link should: 1) Be independently verifiable as actually being from the artist in question.
+                      2) Include an unambiguous reference to the image in question, or an image set that clearly includes the image in question.
+                      3) Include permission to use the images in question for training AI
+     -It is appreciated if members of the community can help reverse-image-search images submitted to the dataset, and help police content.  I am but one goofball, and I 
+          will keep on top of submitted images as best I can, but I don't claim to be magical, omnicient, or even competent at content moderation.   
+
+The second rule of GAI: Use creative, and descriptive names to artstyles instead of artist's names, pen names, trade names, or trademarks from media companies.
+     -This is part of the point of this dataset, to teach the wider art community that AI doesn't just copy existing work.  And also to provide a hedge against possibly
+          litigous activist actions, and future changes to the law.
+     -Users who use AI ethically aren't attempting to counterfeit, displace, or impersonate existing artists.  We want images that have a certain "look" to them, or simply
+          more visually appealing.  So instead of prompting with artist names or existing properties, specify the actual image you want to see, and let the AI do the rest.
+
+The third rule of GAI: NSFW content is permitted, but it is to be kept seperate.  Illegal content like CSAM and "Lightning Rod" content like hate speech will not be 
+      permitted.  Neither will anything else that violates Hugging Face's T.O.s.
+      -What is NSFW? Using Unites States ESRB "Teen" rating, MPAA "PG-13", and typical mass-market social media moderation guidelines are the primary benchmark for what is
+           and isn't NSFW.  If an image straddles the line, discuss.  The primary purpose for separating this content is for appearances.  If something is a bad look for
+           the main dataset, it will get moved.
+      -What is Illegal content? Primarily CSAM, CP, adult material including minors, anything that could get the feds to show up.  AI's are not people, pictures are not 
+           people, but argue about it somewhere else.  This also includes confidential information, medical imaging, or, per rule #1, unauthorized copyrighted material.
+      -What is Lightning Rod content? Hate speech, discriminatory content aimed at groups of people based on heritage, nationality, sex, gender identity, race, ethnicity, 
+           religion, etc.  As well as things like deliberate shock images with no artistic merit.  If an image straddles the line, discuss.
+
+The zero'th rule of GAI: The objective is to build models for Stable Diffusion that work AS WELL or BETTER than models trained on copyrighted material. While using only 
+     AI generated, or volunteered input.  Anything that doesn't serve that goal is subject to moderation and removal.  Don't try to "sneak" copyrighted material in. 
+          Don't try to sneak in illegal, shock, or hateful material.  Don't be a problem.  I can appreciate a good trolling, but we're going to stay on topic here.
+
+          Have fun, and thanks to anyone who wants to contribute! 
+          -DerrangedGadgeteer