Nova Furry AM
A 2d/2.5d furry checkpoint model that can have great details on any type of furs, scales and feathers, which aims to be Anima version of Nova Furry XL
Rules
You cannot use the generated images for commercial use if it's not edited (or just turning it to black and white)
Advertising is always welcome
Recommended settings:
Sampler: Euler a
Steps: 20-30
CFG Scale: 4-6
Prompt: masterpiece, best quality, score_9, score_8, score_7, year 2025, newest, highres, absurdres, very aesthetic, scenery, furry, anthro, {Prompt}
Negative Prompts: human, smooth skin, worst quality, low quality, early, old, score_1, score_2, score_3, cartoon, graphic, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, bad anatomy, bad hands, missing fingers, extra fingers, extra digits, fewer digits, cropped, very displeasing, artist name, blurry, jpeg artifacts, lowres, censor
Description
Changes
Style much closer to Nova Furry XL
Stable prompt adherence
Better E6/R34 adherence
Better coloring
Anima-base v1.0 based
FAQ
Comments (15)
Could you please start to add what major changes can be seen between model versions in the description?
Thank you as always for your work! I just want to preface this by saying I respect your work a great deal; your models have carried me through a lot of image generation. :) I do think we lost some real model range between v2 and v3. Toon or other not explicitly true-to-life forms definitely take a backseat (or sometimes bottom of the trunk space) now as it's pretty difficult to get crisp results in these styles without excessive weighting and the default female curvy form coming through anyway. Looking back, it's actually kind of surprising how flexible v2 was.
I see "Better E6/R34 adherence" in version info. But I tested some very popular e621 tags and the model still doesn't respond to them. Can you please share what your dataset contains, does it contain e621 data? 馃檹
Could you tell me what prompt did you use?
I didn't train the model (just merge) so I really don't know anything about datasets
I can only assume that the artist is expanded and things like long snout works
Hi @Crody, thanks for clarifying. l just use testing prompts with different species, like this dude.
It would be really nice if you can share a list of original models. Many people are looking for e621 finetuned model. 馃檹
@ikekph5聽I shared it as the recipe on our discord
Here, you can join and see it
https://discord.gg/H4jfeDBK6
The recipe which can be loaded using TC-Model-Merge-Planner-GUI can be found in here
https://discord.com/channels/1516698952995442688/1523233043501027459
To view this, first you have to be both Beginner and Member
You can see the tutorial on the welcome page's Pin Messages
@ikekph5聽You gotta understand that its not the creators fault. e621 tags arent officially trained into Anima(its missing e621's dataset entirely). Making loras is a temporary work around. Its like trying to describe colors to a blind person. You can express that the color red is intense, but the blind person wont be able to visualize red.
@LatteLeopard聽I know, that's why I was so hyped when I saw "Better E6/R34 adherence".
I thought it means the model "knows some of e621 tags now", "has been trained with some e621 data/tags". But after some testing I'm confused that non of the e621 tags works, even the most popular ones with hundreds of thousands of images. The model still completely ignores the tags, just as original anima.
It is strange, I use the same prompt from 2.0 and get a differing image on 3.0. Usually more muted and less artsy, any thoughts? I can provide the prompt if needed.
I adjusted the color less vivid than 2.0 since I felt that the color is too much
Provide something like high saturation would make it more vivid
To add to Crody, you can use a color adjustment to your workflow or even use a free software like GIMP to mess with your colors. With artsy stuff some models can be more "strict to the prompt" and using either wildcards for different looks or a lower CFG may help
Didnt work in Easy Diffusion 3.0.16: Error: Could not load the stable-diffusion model! Reason: 'time_embed.0.weight'
Anima isn't supported by easy diffusion I believe. Try swarmui or comfy
Ive put a lot of time into NovaFurryAN, I have some gripes but lets start with positives:
Polish is exceptional. Images take longer to generate, sure, but lines are crisp, fine details are clean, and sharp edges are sharp. Think of it like running an automatic hi-res pass. Takes about as long but it offers a cleaner, sharper output with less loss in detail and artifacting, thank you Anima very cool. It handles longer text prompts better as well if you dont like using tags.
Now for the negatives: It aint fast. Not really a fault of the model though, just architecture. Still worth noting. Tag adhesion is not great. The model responds okay to popular E621 and booru tags but as soon as you stray from the norm youll enter inpainting hell. Its stylistically quite flat, you can pull off some incredibly nice detail but the aformentioned lack of tag support makes that a challenge, again welcome to inpainting hell. The model also has a tendency to lean to more western and cartoony designs. Youll notice things like larger noses, more pronounced muzzles and snouts, simpler and larger fur chunks. It doesnt look bad, but recreating that NovaFurryIL look can be a bit of a challenge at times.
Overall the model is consistent and produces very polished images. If the tag situation can get sorted out and we get some tweaks to the style then Id 100% recommend it. For now, theres plenty of reasons to stick with the IL version, its really personal preference at the moment.
V3.0 works great. Starting with the recommended positive prompt, adding tags, and then also describing the image in natural language is the way to go. It's possible to make complex scenes with multiple characters very consistently. Anima is getting better every day, thanks Crody the GOAT-y for releasing Anima versions of all your excellent checkpoints.