WAIVIER is a Checkpoint Merge of some of my favorite Checkpoints!
Please check the 'About this version' section for more information about what models were used in the merge, at what weights, and what might have changed between versions.
I originally made this just for myself, intending to keep it as a private merge, but as I uploaded more images and posts without having a Checkpoint to link them as resources used, I decided to just upload it so others can use it too.
Recommended Settings:
CFG: 3
Steps: 25 - 40
Sampler: DPM++ 2M SDE
Scheduler: Karras
Clip Skip: 2
VAE Baked-in
I will always recommend using HiRes Fix with all generations to really sharpen things up, but I generate locally so I don't need to budget my buzz for it. It works without the HiRes upscaling process, but the final result will always be cleaner and sharper with it.
This is a learning experience for me, so please have patience with me. I welcome criticism, feedback, requests, and suggestions!
All Rights and permissions are as you'd expect when it comes to a merge. Please generate responsibly.
Description
v1.0 Release. (True Full Release)
Merged XAVIER v1.0, WAI-NSFW-IL-SDXL v14.0, Flax Milk at a (80/15/5) scale
VAE Baked-In
Clip Skip 2
Second iteration of my merge, with slightly heavier leaning toward XAVIER as a base, with a trade-off of less WAI 14 in exchange for a tiny bit of that wonderful Flax influence.
Once the Checkpoint Merge was made, I then merged in just a few of my favorite LoRAs in order to help create that visual style I love.
Used Resources:
Merged locally using SuperMerger extension for Stable Diffusion Forge. A handful of style LoRAs were included in the merging process in order to get the visual style as I like it. There's still much improvement to be made and I look forward to learning more about the process.
Checkpoint is fp16, pruned, and has VAE baked-in. I recommend Clip Skip 2 as that's what I use and that's what it was merged with.
My Recommended Settings:
CFG: 3.0 - 7.0
Steps: 30 - 40
Sampler: DPM++ 2M SDE
Scheduler: Karras
Clip Skip: 2
Please keep in mind these settings tend to work best for me. That does not mean they will work best for you. Experiment and have fun.
FAQ
Comments (21)
The model looks good, but I don't know why. Every time I use it to generate images, they always turn completely black in the last few seconds of generation. I've used the recommended parameters, but this still happens.
Now that's odd. What program do you use locally? Is it setup and able to run other Illustrious models? Do you experience this same issue with other Illustrious models, or is it just mine?
It's hard to troubleshoot exactly what's causing that, since everyone's systems are different, so it's a bit tricky.
Feel free to DM me with the exact prompt you're trying to generate, along with what parameters you're using, and I'll give it a look to see if it's a repeatable error.
This is just a normal merge of some well known and well used Checkpoints, so there's nothing in it that should be causing any issues, but again, everyone's systems and files are different so I'm not going to rule out some weird conflict haha.
@MysticFrostwind Thank you for your response. When I use the WebUI, the black image issue still persists. However, I tried the same recommended parameters in ComfyUI, and the model works fine there. I suspect the problem might be with the WebUI. 🧐
@GentleG I use Forge myself, which is a fork of the WebUI if I remember right.
Glad it works for you in ComfyUI, though!
@GentleG Also, feel free to upload whatever images you made using this Checkpoint merge to the public gallery! I know most of my own stuff is yaoi oriented, I'm fine with anything really, I won't curate (maybe unless it's endless nonstop body horror slop haha).
In the relatively near future I may allow the model to be bid on, and enable on-site generation, I just wanted to be respectful to the creator of the XAVIER merge and avoid any competition between the models.
Think I'll be working toward a v2 with some slightly more finetuning, especially for some added concept and poses that the model currently needs a LoRA to help with.
So yeah, don't feel that you need to upload nsfw content, or yaoi/gay content or nothing at all, I welcome whatever you want to showcase! :D
@MysticFrostwindI think I've finally pinpointed the issue — it lies in the WebUI's sampling precision. After I enabled upscaling, the model started working normally again.
the model is 100% underrated.
I really appreciate the kind words! And I also appreciate you using my Checkpoint to make some of your own content, they look great!
excellet job!!! I'm waiting for the new version.
Link to my comment regarding a version 2 - https://civitai.com/models/1935667?dialog=commentThread&commentId=1127066
TLDR;
Not a priority but absolutely possible if enough people request it / and if people have found any sore-spots that could be improved with a v2 update.
Update released as WAIVIER v2.
@MysticFrostwind Congratulation!
will the model be updated?
Beyond just some basic upkeeping and quality improvement passes, I haven't found much of a reason to force an update so far. I've got some things planned, and I've tested around with private updates and merges, but I've found myself always coming back to v1.
Plus I can only keep one version up for on-site generation, and even then only because I receive yellow buzz for people using the model. Really the model could use with an update with characters from the last few months of anime, but beyond that there's nothing revolutionary in the Illustrious/SDXL community that would really warrant a new version.
I could, of course, be convinced into it if there's a particular issue you've had with v1 that you'd like to see fixed and improved for future versions, so it's not like it's impossible. Models 6 months old now, but as far as IL goes it hasn't really shown it's age yet or any wear-and-tear, honestly.
So yeah, that's how I stand on it currently. I need to investigate WAI-16 to see if I like it's results as much or more than WAI-14 which it was trained with. Beyond that the other two main Checkpoint sources are XAVIER and Flax, which are both older than this model without having any updates either.
Let me know if there's any problems or requests you've found/have that I could look into for a v2.
@MysticFrostwind I understand, I do like the model and think i'd love to try some merges of those same models, I noticed some nice updates or variations exist, I personally prefer WAI-15, and Plant Milk released Walnut which looks awesome and is extremely popular, Xavier also has a r34 version which incorporates extra tags and stuff. I might investigate different percentages of merge and see if I can fine-tune a version that I like for my use-case. Site-wise i understand why it's not worth the further updates tho, but maybe try some of those models if you'd like to experiment with the merge :)
@BobbyKnobby Yeah, when I tried WAI-15 I wasn't really happy with it, so I went back to merging with 14, and liked the results better. Will have to see how 16 performs from a personal level. I make merges based on my own preferences, since if I'm not willing to use my own model then how can I expect anyone else to? lol
By the way, Walnut is 9 months old, it's not new by any stretch. Plant Milk is a collection of different visual styles, not a progression of versions like v1, v2, v3, etc etc. I tested Walnut in my merges before but was more satisfied with the tiny influence from Flax that's used in v1. The entire Plant Milk model variations are all 9 months old, actually. They were published at the same time as alternatives to each other, not version progressions of the same model.
I can definitely try merging and training in some new r34 tags like Xavier's r34 did, so that's a possibility. I don't want to just steal r12x5s's methods though, so it would feel weird to make a new version of waivier based around merging in the same training data he used for his r34 release lol.
I also have to be careful with what sources I use, since xavier was merged using Hyphoria, HomoSimile, and Mol_Keun Mix, I don't want to remerge those back in, or anything that's too derivative of them, since too much weights/influence can also have an adverse effect on the final result.
@MysticFrostwind yeah, that's why I said updates and variations ,not versions. I'm aware walnut isnt an update, i was refering more to wai 15 as an update to 14 and walnut and xavier r34 as variations of the models that you could use. Also, Plant Milk's variations actually released from january to may 2025.
But all that aside, I was just asking if you're experimenting with more models that's it.
in the case of XAVIER r34 i find it just handles certain... "Poses" better. but it's definitely a use-case thing and again, not even for the sake of updating on civit, just suggesting to feed curiosity.
Cheers.
@BobbyKnobby Quick response :)
"wai 15 as an update to 14" - I had a lot of issues with consistent digits when testing 15, but have heard 16 is an overall improvement in that regard. I'd probably skip 15 and instead focus on an updated merge with 16, if I'm going to push for a v2, that is. Glad you've had good results with wai 15 though, if it works for you then enjoy haha.
"xavier r34 as variations of the models that you could use" - Fair, I just don't want r12x5s (user who merged and published xavier) to feel like I'm just copying his homework and putting my name on it. waivier v1 was originally a private merge I made that includes various public and private loras of mine that really push it to the visual style I wanted. I don't want to lose that style with a v2, so I want any updates to be focused entirely on pose improvements and concept improvements rather than changing the overall visual style.
"Plant Milk's variations actually released from january to may 2025" - Going by the release age of the models themselves, IE the pruned fp16 checkpoints, they were released / uploaded 9 months back. I tried a merge with Walnut instead of Flax before, and while results weren't terrible, I didn't like them as much. You're welcome to merge xavier, wai 15, and walnut yourself though, as everyone has different style preferences.
"But all that aside, I was just asking if you're experimenting with more models that's it." - Definitely! Also no worries, I'm long-winded when I write, so don't think I'm mad or defensive haha. I've been considering releasing a new 'Mystic's Wonders' model, and by doing that I can experiment much more with styles than I could with a v2 of waivier.
"in the case of XAVIER r34 i find it just handles certain... "Poses" better." - Yeah, the r34 version of xavier was merged with a huge amount of rule 34 based tags and poses. I haven't ruled it out, just taking my time to figure out what gets the best results for me.
"just suggesting to feed curiosity." - I welcome suggestions and feedback! Again I'm just long-winded haha.
I tried my best to reach out to the creators of the merges I used in the creation of v1, so any v2 I'd release I would like to ensure I'm not stepping on any toes. It's why I waited a few months after v1's release to enable on-site generation and started bidding for it. I didn't want the original xavier to be overshadowed or have to compete with my waivier model.
I also am not the biggest fan of updating a model just for the sake of it, truthfully. IL models haven't progressed by too much in the half a year that waivier was released, and forcing an update when using a single pose/concept lora would get you the same results just feels like wasting storage space from civitai lol.
But again, that's just how I handle things, and I'm not trying to claim my way is the correct way, it's just what I find works best for my own use case. :)
"I don't know too much about training models, but I found that some LoRAs have good performance. What do you think about them?"
I think you could foucus on the part of NSFW by improving position and express and so on. give up some tags which not related to humen. after all FOR sfw model, too much model can do it now . for me , I use banan2 and Mystic's Wonder work togother now, nanobanan has enough power on SFW, open source model is hard to compite
@lighte @BobbyKnobby I pushed a v2 out recently. Should still have a similar visual style to my v1, but with (hopefully) expanded support for poses, newer characters, etc etc. It's not perfect, but after several iterations that I've privately tested, I'm most satisfied with what's now v2.



















