Highly improved version of Prefect-Furry, V6 because there were many in between iterations that were not a substantial increase in quality.
This merger is very capable and can handle many different styles through prompt alone, such as realism, cartoon, and anime.
All images shown in the preview for Prefect-Furry-V6 were NOT post processed! No Hi-Res, no A Detailer, pure output.
CFG scale of 9 is recommended, as it improves cohesion over 7 by a small amount, but noticeable.
If using Automatic1111, here are the recommended styles I use in all generations that make a huge difference:
Prompt:
masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, depth of field,
Negative prompt:
old, oldest, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digits, fewer digits, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, conjoined, bad ai-generated, Patreon, logo, monochrome, fuzzy details, paint strokes,
Description
Merging WaiNSFW v15 and Nova Furry XL Illustrious v6b
FAQ
Comments (2)
Possibly the best current merge on the site right now. After doing extensive testing with this in conjunction with against both Nova and WAI-NSFW in scenarios:
1) Nova and WAI only performed better in three specific edge cases, otherwise, style adherence and prompts were uniformly better than both: One, if the artist was hard-baked or overrepresented across the both of them (Houtengeki, for instance), it tended to overbake faces slightly, which is easily ameliorated by ADetailer or lower LoRA strength. Two, pixel art - this merge simply cannot do pixel art / low poly as well as WAI or other checkpoints. Three, dramatic lighting / high contrast needs to be given a higher strength or more prompts compared to either Nova or WAI.
2) The number of accidental / non-sense hallucinations (extra fingers, toes, ghost hands, etc.) seemed to be lower in this merge with minimal prompting compared to both Nova and WAI on their own.
3) Prompts are more closely followed and conflicting tags (ie: two articles of identical clothing in different colors) do not try to merge as much compared to the standalone checkpoints.
4) The only LoRAs in my library that have failed to generate or make a distinctive, non-checkpoint style are those that are tailored for very specific checkpoints and even then, 99% of them work perfectly, if not better than on independent checkpoints.
5) There is a slight bias towards generating a small amount of pubic hair by default as well, but this is easily solved by putting it into the negative prompts.
In my opinion, this is a perfect, generalist checkpoint that will suit the vast majority of prompting for the average user.
Thank you for the amazing breakdown! I will work in the future to try to resolve the shortcomings in future updates as quality is my highest priority with lowest effort needed to achieve it
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.