v1.0 Anima preview, finetuned at 1024px
Usage Tips:
Positive Prompt:
masterpiece, best quality, anime coloringNegative Prompt: score_1, score_2, score_3, blurry, jpeg artifacts,lowres, worst quality, low quality, bad hands, bad anatomy,signature, text, watermark, censor, bar censor, mosaic censorship, mosaic censor, sound effects, blubble speech, patreon logo, patreon username, twitter username, twitter logo, chibi, nsfw, explicit, blur, multiple views or similars.
Technical Settings:
Sampler: Euler a, Euler, ER_SDE
Steps: 20-28
Scheduler: Simple, Normal, Beta
CFG Scale: 4.5-6.0
AuraFlowShift: 3.0 - 5.0
Resolution: 1024x1024 or any aspect ratio that falls within ~~1.048.576 pixels.
v4.0 a bit more stable, a bit more leaning into medium boobs, more neutral anime style in general, use the baked-in VAE, should get pretty degenerate as this was trained over v-pred 1.0 if vpred can do it, this should be able to do it.
v3.6 does lean less into big booba in most prompts, is more 'reserved' in the default size as a plus is able to get far darker in some prompts compared to previous versions with the usual caveat of being more noisy since anime in dark scenes is ultra noisy, in some prompts the anime style is stronger, this version likes Euler Dy CFG++ and Euler A Dy CFG++ making gens smoother more like anime itself, has Anzhc's B3 Decoder VAE baked in, DO NOT USE OTHER VAEs.
v3.0 now more stable in general.
Custom mix of finetuned styles on NoobAI v-pred to reach a simple aesthetic I liked, focused on 2D like anime without looking like an specific anime all the time, and clear outlines, despite some tokens still bleeding styles in NoobAi, its decently consistent, this mix has some of @r3c's r3mix in it whatever they did to it helped a lot + a bunch of finetuned styles with train diff from NoobAI v-pred 1.0.
Most Illust 0.1 and Noob 1.0 Lora/dora/glora/lycoris/loha will work but the quality of the denoising will vary, ymmv.
Usage Tips:
Positive Prompt:
masterpiece, best qualityNegative Prompt:
worst quality, low quality, worst aesthetic, old, early, blurry, lowres, signature, artist name, watermark, twitter username, sketch, logo, furry, text, speech bubble, censored, ai-generated, censorship, censor, mosaic censoror similars as needed
Technical Settings:
Sampler: Euler a, Euler, Euler DY CFG++, Euler A DY CFG++
Steps: 20-28
Scheduler: Simple, Normal
CFG Scale: 4.5-6.0
CFG++ Scale: 1.0 - 2.0
Resolution: 1024x1024 or any aspect ratio that falls within ~~1.048.576 pixels.
Remember to set these in your reforge for best results.
If you want to support me, consider liking the model, and/or tipping buzz/donating to my Ko-fi, as this was finetuned on 2x4090 over 435k steps across all styles.
Description
FAQ
Comments (11)
This model is great and I loved the artstyle and it's strength to make both humans and furries.
By the way I used 896x1152 with 17 steps DPM++ 2M - SGM UNIFORM with CFG SCALE 5.0 for my image generation.
Let me know if there are other settings you have used and worked great for your WebUI you are using.
NOTE: I'm using ComfyUI so workflows can vary for everyone.
@libright I prefer Euler, Simple or Normal with CFG5
@bluvoll I would love to hear how others have experimented with their setup and sharing their results here or in another discussion. The gallery is great and can expand it with furries to show it's capable and versatile too, as I have seen it on my end.
@libright I think that a conversation won't happen, but I'm glad you liked the ckpt, feel free to post samples, as I don't delete furry even if its not my cup of tea.
@bluvoll i will have some ideas drafted soon, so expect some great expansion on it's galleries.
I just prefer Euler/Euler A for anime models.
one of the best model with pure anime style, only thing i can recomend is for next version add more medium small size breast character in the mix, it tends to do all female with huge breasts mostly, can be fixed with negative prompt but would be better overall if model does not always pick large/huge. apart from that model is perfect, clean anime style.
@MrFlex I can try if I make a v4, but no promises of me even doing it.
@MrFlex try v3.6 should be less big booba, but prompts did change a fair bit.
@bluvoll its not that i want less but its good for variation, i will try it out now looks amazing from sample images
@MrFlex Yes, I get it, more variance, less 'default' to that size, I get you.
Details
Files
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.