<h1 id="the-style-of-mangamaster.">The style of Mangamaster.</h1>I made this lora to try and reproduce the style of Mangamaster.I added a new version trained of Illustrious 2+. I also changed the activation tag to <code>mangamaster</code>This LoRa is trained on Pony V6 and Illustrious 0.1, but I find the results in Autismmix (Pony) and Obsession (Illustrious) more to my liking. This lora works great in both checkpoints!Both image sets were made with Addetailer extension enabled and using a face and eye model (and some other were needed).Trigger: <code>Mangamast3r,</code>Advised strength: 0.7-0.8Prompting IllustriousFor all versions I recommend you use: <code>masterpiece, best quality, amazing quality, very aesthetic, high resolution,</code> in your prompt and I myself am using the following negative prompt: <code>graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, very displeasing, lowres, (worst quality, bad quality:1.2), bad anatomy, jpeg artifacts, signature, watermark, username, signature, simple background, conjoined,ai-generated, muscular, 3D, polaroid,</code>Prompting PonyFor all version I recommend you use: <code>score_9, score_8_up, score_7_up, score_6_up,</code> in your prompt and <code>score_4, score_5, censored, mutated, bad hands, deformed hands, missing fingers, multiple fingers, multiple digits, extra digits,</code> in your negatives. It will greatly improve the results. Version 1: This is one of the first lora I have made and trained on 6000 steps and 10 epochs. The dataset was a of normal quality. The images created by this style have a smooth feel to it. I used the addetailer extension to help with the faces and eyesVersion 2: Trained on a dataset double that of version 1, but with a lower step count (4100 steps to be precise). It's feel's stronger in style then version 1. I used the addetailer extension to help with the faces and eyesVersion 3: Trained on a dataset of 30 images of high resolution. The number of steps has been increased to 4500. The style is very strong so giving it a lower weight might help. This version produces sharp lines. Also the trigger word got changed to <code>Mangamast3r</code>Thank you for reading and have fun!

Mangamaster Style - Pony / Illustrious

This is a new version with a completly rebuild (and bigger) dataset.I made this after noticing that I overtrained the 1.1 version a bit to much.

Files with SHA256: f2706441fe9e29c68589cb54b181b4bd516dbe78f25a483591f45f27d4be5c23