Version XVII
This is an updated personal Illustrious merge, consisting of some of my favorite checkpoints and LoRA's. Bringing it in to 2026! All and massive credit goes to the original creators of the checkpoints and LoRA's.
Tip: Use and weigh for great fur: realistic, (fluffy, detailed fur:1.4),
Improvements compared to Phased Merge XIII:
Even better and more detailed fur.
More flexible and dynamic poses.
Better and more flexible expressions.
Brighter.
More coherent and more detailed overall.
Things that possibly got a little worse, even though LoRAs were weighted a little less than the previous merges:
Species variety in some cases.
Hands and paws in some cases.
Teeth.
The biggest improvement is that images generated with this merge can now be used commercially! Specifically some of the LoRA's were reselected with this in mind. Version XIII and VII still use restricted LoRA's, so no selling images generated from the older versions, only with XVII.
Models used for this merge:
Nova Animal V10: https://civarchive.com/models/784543?modelVersionId=2436591
Amanatsu V2: https://civarchive.com/models/1325426?modelVersionId=2288934
Tiger Milk 🐯🌿: https://civarchive.com/models/1635490?modelVersionId=1851225
LoRA's used:
Adjot / Snowskau style: https://civarchive.com/models/1500955/adjot-snowskau-artist-style-illustrious
Zaush Art Style: https://civarchive.com/models/1076338/zaush-artstyle-ponyillustrious
Furry Enhancer (/w permission, tysm!<3): https://civarchive.com/models/310964?modelVersionId=1780880
Detail Tweaker XL: https://civarchive.com/models/122359/detail-tweaker-xl?modelVersionId=135867
Showcase images:
All images in the showcase have been made with 1.5x latent upscale, automatic face and eye detailing, and ESRGAN upscale. Made in ComfyUI in one automatic pipeline. So all in all, be aware that 'base/raw' gens look less refined. Even with the ComfyUI pipeline, inpainting would be required often for paws, teeth and hands to make things perfect. Images have been cherry-picked a little to show off the enhanced fur, details and posing. Gens are based on older prompts of mine.
What does this merge intend to do?
It's merged to change the style of the base Nova Animal closer to my personal preference, with a focus on good fur detailing. It changes the styles of the faces, fur, details and backgrounds compared to the base models. I tried to keep a somewhat cartoony/expressive vibe to it as well.
Nova Animal is mixed with Amanatsu (anime checkpoint) for more flexibility, and Tiger Milk 🐯🌿for a little extra flexibility and further fur detailing. The LoRA's help that too, especially the Furry Enhancer LoRA is quite incredible. The Snowskau LoRA directs the fur style a little, and the Zaush LoRA is lightly weighted to give an aesthetic to faces I personally enjoy. The Zaush LoRA might make other faces (rodents for example) less like they should be though, so a LoRA for that might be needed in those cases.
I did weigh the LoRA's a little less this time around, so the model should be a little more flexible.
License and usage:
This merge honors the license settings the checkpoint/LoRA creators had set for their work. I received permission from the Furry Enhancer LoRA creator to use it in the merge. With this new Phased Merge VXVII you can now commercially use the generated outputs as well! The older versions XIII and VII are still commercially restricted due to the LoRA's used in those versions!
I've only required creator credit in the license in this merge, so that the original creators get properly accredited, this merge is all thanks to them!
Recommended settings:
Resolution: I recommend genning at 1536x1024 (or reversed) for best details. 1280x1280 is possible too of course, even things like 1728x768. If VRAM is a concern or you get wonky results, step down to 'old' resolutions like 1280x768 and upscale more and so on.
Sampler: Euler A for best performance, though other samplers work too of course, each one having some/slight differences. It's personal preference.
Steps: 20-50 steps (30+ preferred imo)
CFG: 3-5
VAE: Integrated. Illustrious-XL-v2.0: https://civarchive.com/models/1684460/illustrious-xl-v20
I can wholeheartedly recommend using Hires Fix if you're an on site user.
'Recommended' base prompts:
Prompt: masterpiece, best quality, very ava, amazing quality, very aesthetic, ultra-detailed, 4k, highres, volumetric lighting, ray tracing, realistic,Negative Prompt: kid, cub, loli, child, bad quality, worst quality, worst detail, lowres, error, blurry, unfinished, displeasing, oldest, early,These can be tailored to personal preferences. Many people are divided on the 'best' prompts, and I don't know which are objectively best. This is mostly what I've been using and works for me fairly well. So play around for sure! Results vary.
Since the merge is based on a realism model, you will have to weigh some things a bit more to get the results you want. For a furred kobold I had to weigh things heavily for example. Proper prompts are important!
Important: If you want truly detailed, fluffy fur in most cases, put in and weigh to personal preference:
Prompt: realistic, (fluffy, detailed fur:1.4),Workflow:
Finally, if you download the RAW showcase images, you can drag it into ComfyUI to get my workflow. It's probably inefficient, and am quite sure it's not as good as others do it, but it could be a decent starting point for you. The workflow regularly hits up to almost 20gb of VRAM though with 1280x896 res because of the upscaling and detailing, and genning batches of 4, so you'd need a card able to manage that. Or step down resolution, cut nodes, and optimize the heck out of ComfyUI, idk. I honestly recommend ReForge for example if you start genning locally. It's easier, has some great plugins, and especially inpainting is quite handy in there. I use Krita with AI diffusion plugin since I prefer ComfyUI, but you can't go wrong with ReForge or similar updated webui's.
You should also even be able to drag the Phased Merge checkpoints itself into ComfyUI and see the simple merge workflow. It's super simple, and for sure not on the amazing level of what the Nova or Furry Enhancer creator or many others are doing! I'm just a guy fiddling with buttons hehe.
Hope you enjoy! Have fun genning, make the best of it, and take care of yourself! ^^
Version XIII
This is a personal Illustrious merge, consisting of some of my favorite checkpoints and LoRA's. All and massive credit goes to the original creators!
Models used for this merge:
Nova Animal V5a: https://civarchive.com/models/784543?modelVersionId=1936160
Amanatsu: https://civarchive.com/models/1325426?modelVersionId=1496755
Tiger Milk 🐯🌿: https://civarchive.com/models/1635490?modelVersionId=1851225
LoRA's used:
Snowskau style: https://civarchive.com/models/1178677/snowskau-style-illustrious
Zaush Artist Style: https://civarchive.com/models/1488390/zaush-artist-style-illustrious
Furry Enhancer (/w permission, tysm!<3): https://civarchive.com/models/310964/furry-enhancer
Detail Tweaker XL: https://civarchive.com/models/122359/detail-tweaker-xl?modelVersionId=135867
Showcase images:
All images in the showcase have been made with Ultimate SD Upscale x2 (Foolhardy Remacri), face and eye detailing, and small refining passes. Made in ComfyUI in one automatic pipeline. So all in all, be aware that 'base/raw' gens look a little less refined, though it's surprisingly close. Images have been cherry-picked a bit more than in my previous merge to show off the enhanced fur, details and posing. Gens are based on older prompts of mine.
What does this merge intend to do?
It's merged to change the style of the base Nova Animal closer to my personal preference, which is in between cartoony and realistic with good fur detailing. It changes the styles of the faces, fur, details and backgrounds compared to the base models.
Nova Animal is mixed with Amanatsu (anime checkpoint) for more flexibility, and Tiger Milk 🐯🌿for a little extra flexibility and further fur detailing. The LoRA's help that too, especially the Furry Enhancer LoRA is quite incredible. Be sure to check out that creator's work, because it's fantastic! The Snowskau LoRA directs the fur style, and the Zaush LoRA is lightly weighted to give an aesthetic to faces I personally enjoy. The Zaush LoRA might make other faces (rodents for example) less like they should be though, so a LoRA for that might be needed in those cases.
Hands and fingers are fairly consistent as far as AI goes, on the level as the base models.
License and usage:
This merge honors the license settings the checkpoint/LoRA creators had set for their work. I received permission from the Furry Enhancer LoRA creator to use it in the merge. This merge is not intended for commercial purposes, only to freely gen cool stuff! <3
(Due to CivitAI Model Page limitation, I cannot set different licenses for each version. This version, VII, is not intended for commercial use!)
I've only required creator credit in the settings so that the original creators get properly accredited, this merge is all thanks to them!
Recommended settings:
Resolution: I recommend genning at 1536x1024 (or reversed) for best details. 1280x1280 is possible too of course, even things like 1728x768. If VRAM is a concern, step down to 'old' resolutions like 1280x768 and so on.
Sampler: Euler A for best performance, though other samplers work too of course, each one having some/slight differences. It's personal preference.
Steps: 20-50 steps (30+ preferred imo)
CFG: 3-5
VAE: Integrated
'Recommended' base prompts:
Prompt: masterpiece, best quality, very ava, amazing quality, very aesthetic, ultra-detailed, 4k, highres, volumetric lighting, ray tracing, realistic,Negative Prompt: kid, cub, loli, child, bad quality, worst quality, worst detail, lowres, error, blurry, unfinished, displeasing, oldest, early,These can be tailored to personal preferences. Many people are divided on the 'best' prompts, and I don't know which are objectively best. This is mostly what I've been using and works for me fairly well. So play around for sure! Results vary.
Since the merge is based on a realism model, you might have to weigh some things a bit more to get the results you want. Proper prompts are important!
Important: If you want truly detailed, fluffy fur in most cases, put in and weigh to personal preference:
Prompt: realistic, (fluffy, detailed fur:1.4),Workflow:
Finally, if you download the RAW images, you can drag it into ComfyUI to get my workflow. It's probably inefficient, and am quite sure it's not as good as others do it, but it could be a decent starting point for you. The workflow regularly hits up to almost 20gb of VRAM though with 1536x1024 res, so you'd need a card able to manage that. Or step down resolution, cut nodes, and optimize the heck out of ComfyUI, idk.
You should even be able to drag the checkpoint itself into ComfyUI and see the simple merge workflow. It's super simple, and for sure not on the amazing level of what the Nova creator or many others are doing! I'm just a guy turning buttons hehe.
Hope you enjoy! Have fun genning, make the best of it, and take care of yourself! ^^
Version VII
This is a personal Illustrious merge with a touch of Pony realism, consisting of some of my favorite checkpoints and LoRA's. All and massive credit goes to the original creators!
Models used for this merge:
Nova Furry V8a: https://civarchive.com/models/503815?modelVersionId=1865201
Nova Animal V4a: https://civarchive.com/models/784543/nova-animal-xl
Pony Realism v22: https://civarchive.com/models/372465?modelVersionId=914390
LoRA's used:
Incase + gothic style mix: https://civarchive.com/models/1059581/incase-gothic-style-mix-or-illustrious
Snowskau style: https://civarchive.com/models/1178677/snowskau-style-illustrious
Detail Tweaker XL: https://civarchive.com/models/122359/detail-tweaker-xl?modelVersionId=135867
XL More Artful: https://civarchive.com/models/124347/xlmoreart-full-xlreal-enhancer?modelVersionId=152309
Showcase images:
All images in the showcase have been made with latent x2 latent upscaling (Foolhardy Remacri), face and eye detailing (though face detailing can be considered a relic, maybe, possibly. Eye detailing is mint though IMO), and small refining passes. Made with ComfyUI. So all in all, be aware that 'base/raw' gens look less refined. Images have been 'cherry-picked' from a single batch of 4 images total, gens based on older prompts of mine.
What does this merge intend to do?
It's merged to change the style of the base Nova Furry a bit to my personal preference, and so I can post images to it correctly lol, since I don't want to flood the Nova pages with unrealistic images to the checkpoint. Anyway, it changes the faces, fur, details and backgrounds a smidge. Nova Furry is mixed with Nova Animal and a touch of Pony Realism for a little more fur detail without going into the true realism zones. The LoRA's help that too. Consider this a somewhat 'in between' checkpoint, with my preferred style added in. The changes can usually be seen in the faces, fur detailing, as well as the backgrounds which are leaning more towards photographic backgrounds now.
Hands and fingers are fairly consistent as far as AI goes, but it's not better than the base models. In fact, fingers and nails might be slightly worse due to the merge, because normal Nova Furry is mixed with realism models, and might require a touch more inpainting sometimes for those very detail oriented.
Note that due to merging, the model has become a little less flexible than the base checkpoints. Also due to Pony Realism being used on top of the Nova models, it's possible that it's limiting the pose diversity and expressions as well, it's a Pony Realism thing. It's not that heavily used though, so it's effect shouldn't be that pronounced.
Edit: After genning a bit more, it is indeed less dynamic/expressive. Define and weight the prompts a bit more to compensate.
License and usage:
Since it's mostly based on Nova and the creator doesn't want the model to be used for commercial purposes, this model honors that as well. It is not intended for commercial purposes, only to freely gen cool stuff! <3
(Due to CivitAI Model Page limitation, I cannot set different licenses for each version. This version, VII, is not intended for commercial use!)
I've only required creator credit in the settings so that the original creators get properly accredited, this merge is all thanks to them!
Recommended settings:
Resolution: I recommend genning at 1536x1024 (or reversed) for best details. 1280x1280 is possible too of course, even things like 1728x768. If VRAM is a concern, step down to 'old' resolutions like 1280x768 and so on.
Sampler: Euler A for best performance, though other samplers work too of course. It's personal preference.
Steps: 20-50 steps (30+ recommended imo)
CFG: 3-5
VAE: Integrated
'Recommended' base prompts:
Prompt: masterpiece, best quality, very ava, amazing quality, very aesthetic, ultra-detailed, 4k, highres, volumetric lighting, ray tracing, realistic,
Negative Prompt: kid, cub, loli, child, bad quality, worst quality, worst detail, lowres, error, unfinished, displeasing, oldest, early, username, watermark, signature,
These can be tailored to personal preferences. Many people are divided on the 'best' prompts, and I don't know which are objectively best. This is mostly what I've been using and works for me fairly well. So play around for sure! Results vary. Keep 'Realistic' in if you want a bit more fur fidelity though.
Workflow:
Finally, if you download the RAW images, you can drag it into ComfyUI to get my workflow. It's probably inefficient, and am quite sure it's not as good as others do it, but it could be a decent starting point for you. The workflow regularly hits up to almost 20gb of VRAM though with 1536x1024 res, so you'd need a card able to manage that. Or step down resolution, cut nodes, and optimize the heck out of ComfyUI, idk.
You should even be able to drag the checkpoint itself into ComfyUI and see the simple merge workflow. It's super simple, and for sure not on the amazing level of what the Nova creator is doing!
Hope you enjoy! Have fun genning, make the best of it, and take care of yourself! ^^
Description
Personal merge. See page for all the details.
FAQ
Comments (29)
You beautiful bastard, you did it!
How dare you accuse this fine gentleman of being born out of wedlock!
Awesome model! Really love the style it gives.
Thank you so much! It means a lot coming from someone that knows what they're doing ^^ It's only a simple merge. I saw your Stormtooth XL merge too, looks amazing! I've been eyeing the plant milk models too, love that they're so creative. It's kinda funny and coincidental that we basically made a more realistic merge at the same time, though we have a different focus. Oh, and I've seen the Bananasplitz model around too, loved that too! Anyway, thanks again! <3
@Aereas youre very welcome 😂
I appreciate the feedback on my models too.
For a future update, I think the model AnthroMix would fit amazingly with this model.
Blind guess for a good result
DARE @ 0.55 weight.
or
Weighted Sum @ 0.45
(Assuming model B is AnthroMix)
Btw how did you mix pony model into this?
train difference or DARE?
@LatteLeopard I wish I could give you a good technical explanation, but I just simply ran it through the extremely basic ComfyUI merging workflow with some LoRA's added in, just fiddling with the weights until I got something that seemed okay, hehe. No clue how they handle it on the backend, and how Pony realism succesfully got mixed in. There's some signs it worked though I think. I really am just someone bumbling about, turning knobs here and there. Anyways, I used this, it's done in minutes, don't really know more about it xD -> https://comfyui-wiki.com/en/workflows/model-merging
And thanks for the tip, AnthroMix indeed looks amazing! I'm gonna browse around and keep my eyes open for models that could make it even better, so I added it to the list. ^^ I really want more pronounced fur, without too much realism and sacrificing pose/expression flexibility. Seems a delicate thing to manage.
@Aereas its okay, A technical explination would have totally gone over my head anyway.
But thank you for the link to the nodes/workflow. Im sure Ill try it out one day, but I currently only use comfyui for Wan 2.1. Im stupidly faithful to Automatic1111's extensions Supermerger and Untitled merger.
If your wondering why im shocked you merged them. its because they have a moderate difference in unet
(which is something I totally recommend you learn. its easy as shit. it just looks complicated.)
I learned everything from this.
https://rentry.co/BlockMergeExplain
And I was shocked with how simple it was, because the idea of block wieght merging seemed like a 200IQ kinda thing.
@LatteLeopard Thank you very much for the information! I'll definitely start learning it if/when I want to get more in depth with merging hehe. That guide does indeed look very, very understandable and easy to follow, so I definitely bookmarked it for later.
As for the merge, I think Pony actually did have a negative effect after genning a lot more with it. Yes it worked, sort of, but... Gens can look a bit smudgy, with some weird artifacting. Plus the fur isn't as good as it can be. Upscaled it mostly looks alright(?), but doesn't really feel 100% to me yet.
I'm trying some other things, and might release a way better version soon... ^^
1280x1280 OutOfMemoryError. On PlantMilk does not give such an error. I have 12 GB of video memory. WHY?
Good question, I don't know. The model size is the same (it's the 'normal' size for XL models). What program/workflow are you using?
I'd recommend checking how much VRAM you're using exactly when using Plant milk, then switching to this model, and checking if it behaves that differently.
I don't think it should behave different than Plant milk models, but it is my first (and simple) merge, maybe it behaves differently because of the models/LoRA's that are integrated? Idk
@Aereas I use Stable Diffusion by AUTOMATIC1111.
On PlantMilk_flax:
Time taken: 1 min. 33.6 sec.
A: 15.83 GB, R: 22.13 GB, Sys: 12.0/11.9844 GB (100.0%)
On PhasedMerge:
OutOfMemoryError: CUDA out of memory. Tried to allocate 14.42 GiB. GPU
Time taken: 1 min. 41.2 sec.
A: 15.99 GB, R: 22.11 GB, Sys: 12.0/11.9844 GB (100.0%)
@Sofia_Dragoness Hmm I can't say things with 100% certainty, but here are some thoughts:
1. Automatic1111 is quite outdated, it hasn't been maintained and updated for a while. Consider switching over to Forge or SwarmUI. They're faster and better optimized too, and work with more models. It might solve the issue. Forge has the exact same UI as Automatic 1111, so that might be a great option.
2. Increase the system's virtual memory / page file a little bit. It might be that it just needs a little boost to completely process the model.
3. Not a real solution, but you can also consider stepping down resolutions a tiny bit to for example 1280x1024, and then upscaling more. Could be a workaround.
It is strange that this merge gives you an out of memory error, and the other model with the exact same size doesn't. The differences in RAM usage seem really small too, but maybe it's just enough to give you that error.
Wish I could help you more, but maybe the thoughts/tips above will do the trick!
@Aereas 1. Okay, but I have a AMD video card.
2.Okay, I try it.
3. It is do not working.
@Sofia_Dragoness Hey, I released a new merge, that might work for you. In the old merge, I included Pony Realism, which had a different architecture from the other Illustrious models. It might've been the reason as to why it used more VRAM. I'm not sure though, it's a guess, but it's worth trying the new one if you like this kind of style!
@Aereas Thanks! Xlll version is work on 1300x1300 and 1600x1600! I like this style, because I try PerfectMerge model a years ago.
@Sofia_Dragoness That's awesome to hear! :D Glad it works now, thanks for trying this merge, and I hope you enjoy it! <3
@Aereas Yes. I wonder if you made a new version so that everything works for me?))
@Sofia_Dragoness Hehe, that was a small part of it. It was mostly because when I was genning more with the previous merge (VII), I noticed some slight smudginess and artifacting at times. Then, LatteLeopard mentioned Pony Realism having a different architecture, so I tried merging only Illustrious models. After posting the merge, I circled back to your comment, realizing that Pony Realism might have been the problem :) So it was mostly a happy coincidence that it works now, and I'm happy the RAM requirements are better now and that you can use it! ^^
Great therianthrope model! Beautiful aesthetic. Can produce fine details like stitching in leather and patterns in fabric. This is a winner!
Thanks for trying it, the kind words, and the amazing tip! Makes me happy to hear that you've got a use for it and are happy with it <3
Really great model for images with more realistic style. I definitely will be trying out what else can I create with it.
This is amazing, just amazing
The Quality of details of the outputs are just simply outstanding!!!
Def one of my new favs! ❤️❤️❤️
Thank you so much, loved seeing your gens! <3<3
I'm probably uploading a newer version tonight, in my opinion it's going to be even better :D More dynamic and flexible, better expressions and poses, better fur/scale detailing. Plus, I noticed this current merge can have some slight artifacting/blurriness in some gens because of Pony Realism that was mixed in, which has a different architecture. That should be rectified with the new merge too, without losing any of the style or fur detailing. Keep an eye on the model page if you're interested, and thanks a bunch again for your kind words, so cool! <3<3
@Aereas oh wow!!! Def ill be waiting for the new version!
Havent encountered artifacts and such (or maybe they were so small that i didnt noticed at all), but an improvement is always welcomed, thanks for your work
You deserve having a cute smile and very much happyness UvU ❤️ ❤️ ❤️
@SXCD2719 Just uploaded the new merge, should be available to try if you're interested. And thank you for the well wishes, it's very, very much appreciated! <3
@Aereas Totally loved it!!!
It feels amazing when we talk about prompt adherence, been using danbooru and e621 tags and everything works perfectly with just a little patience
Amazing work! this is totally a step forward!
Thaaaanks ^v^ ❤❤❤