<h3 id="the-next-iteration-of-the-marefusion-2-models-this-time-based-on-the-wurstchen-architecture-instead-of-sd1">The next iteration of the marefusion 2 models, this time based on the Wurstchen architecture instead of SD1</h3><h2 id="please-read-the-setup-instructions-below">PLEASE READ THE SETUP INSTRUCTIONS BELOW</h2><p></p><p>This model is a finetuned stable cascade stage C 1B model trained on the Marefusion 2.4 curated uncensored dataset of 55k pony images, using a single 2080ti with the 22 GB VRAM mod, and is prompted like how images are tagged on derpibooru (with a few exceptions, I had to change some tags to be more compatible with AI text encoders, i.e. ":p" -&gt; "blep", see the included tags file for details), and with over 130 different style tags that can be mixed and matched, will enable you to find the perfect style to fit your vision, at 1024 resolution, 1536, 2048, anywhere in between, and beyond.</p><p></p><h3 id="about-the-file">ABOUT THE FILE</h3><p>The file is a compressed archive of the stage C and text encoder weights, as well as a file containing all the tags and their frequencies in the dataset, a file with all the artist tags above 50 occurrences, a file of that as a wildcard, and a basic comfyUI workflow to get you started</p><h3 id="setup-instructions">SETUP INSTRUCTIONS</h3><p>This is a wurstchen/stable cascade model, and so it is not just a simple checkpoint file, you must place everything in the correct places.</p><p>you will need the stage C and text encoder weights from here, stage B, stage A, and effnet encoder from the stable cascade huggingface: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-cascade/tree/main">https://huggingface.co/stabilityai/stable-cascade/tree/main</a> or you could get the stage B from sotediffusion: <a target="_blank" rel="ugc" href="https://huggingface.co/Disty0/sotediffusion-v2/tree/main">https://huggingface.co/Disty0/sotediffusion-v2/tree/main</a> which seems to work better than the original stage B model</p><p>For comfyUI, place the stage C and stage B into the unet folder, the text encoder into the clip folder, and the stage A and effnet encoder into the VAE folder, then use the load diffusion model, load CLIP, and load VAE nodes to load them, a basic comfyui workflow is provided in the files</p><p>For other UIs, please consult the readme for any stable cascade support, and how to set it up</p><p></p><p>screenshots of the MLP shows and movies are also included<br />to get the g4 show style, use "my little pony: friendship is magic"<br />to get the pony life style, use "my little pony: pony life"<br />to get the g4 movie style, use "my little pony: the movie"<br />to get the g5 movie style, use "my little pony: a new generation"<br />to get the g5 3D show style, use "my little pony: make your mark"<br />to get the g5 2D show style, use "my little pony: tell your tale"</p><p></p><p>contact me if you want the training data, the dataset is constantly evolving, so I can't just put it online somewhere without it becoming outdated in a week</p><p>discord server --&gt; https://discord.gg/AScpPF7StH &lt;--</p>

Marefusion 2: Emaregence

<p>48 epochs, 24 with text encoder training, 24 without</p>

Files with SHA256: e30276d2a2dbf213888c97321c076baba6fde134f679cbd38d2f0836ebdc42a0