Another Furry Model... woo.
A development branch of a SDXL model that hopes to eventually surpass Alpha Tango in quality and accuracy, uploaded for public use and to gather feedback. If you don't have the hardware or wish to try before you download, you can use this model for free on the sites linked below!
Important information below, continue reading before downloading!
<About the Model>
A furry model designed for accuracy and ease of use with great results attainable with no extra resources.
Can perform a variety of styles (non-artist), with realistic-styled images being the main purpose of the model. I may not have posted many example images but yes, the model does understand a wide variety of styles. It can do more than what is posted..
Wish to use this model but don't have the hardware to run locally?
That's no problem! You can run this model for free on the sites linked below:
(Remember to read and follow the ToS of each site!)
SeaArt.AI
https://www.seaart.ai/models/detail/99962874251be653bc6dbf93828d13b6
Tensor.Art
https://tensor.art/models/795465837986372708
PixAI.Art
https://pixai.art/model/1748518760348069774/1813326239916350421
Version 0.6.5 minor update release, change list below:
2D version
Higher style comprehension (no realism with this one)
Better scenery/background output
Higher image clarity
********************************************************************************************
I used custom merging resources to create this model, I have them posted here:
https://civarchive.com/models/644327?modelVersionId=720766********************************************************************************************
Prompting uses E621 tagging. If you ever used Alpha Tango, Yiffy, Indigo, Redwater, etc. or know how to prompt for SD1.5 models based on E621 then there will be very little adjustments needed for prompting.
Score tags are unnecessary, but you are free to do as you please.
1024x1024 and upwards is a good resolution for generating images with this model, but do as you please.
A great jumping off point for merges or general use.
Use however you please, feedback is appreciated, crediting is unnecessary.
Generation guide linked here (WIP).
This is a Pony (SDXL) model, this is not a Lightning model, nor a natural language model.
CFG at least 7, Steps 30-50, Euler A or DPM++2m (Karras).
If you are receiving artifact/noise errors there is a guide in the suggested resources to help assist.
If you do not have time to read the guide and wish for a quick solution to the problem, then remove any <: 1.2> or multipliers in this format to fix the error. Using {}, [], (), +, -, should still be fine however. This solution should only really be used if the issue occurs.
This error is not model-exclusive and is an issue with SDXL. That doesn't mean you can't use <: *.*> multipliers, just know that this format can, on occasion, cause issues with A1111 if the format is used with large prompts or numerous times.
Currently Planned in Next Version:
Nothing
ETA: TBD.
Please do not reupload on other sites behind a user-placed paywall or credit requirement, this is designed to be used for free.
Description
Interim model released for public use as I continue finetuning three specialized merges.
With this model you get:
Concept comprehension +
Style output - (It's down)
More complex backgrounds.
More anime/pixel art support.
Better detail output.
~~~~~~~~~~~~~~~~~~~~~~
I would recommend using the fp32 version over fp16, as results vary wildly between precisions. But that's just my recommendation, do as you please.
FAQ
Comments (11)
What tags should be used?I see "real e621" What does that mean?
Is there a list?
https://civitai.com/posts/2672445 12.9GB full model
Looks good,but she has human hands
It generates humans or human hands. How do I consistently make a furry?
With the other prompts the lynx reminds me of an Asian woman, do you know anything about that?I don't want Asian women
Is this Pony based? I ask because the NSFW is MUCH better than your usual SDXL models can typically do (unless they're Pony).
However, I'm also finding that I can prompt this in the same way I used to prompt 1.5 models like Redwater, EasyFluff, FluffyUniverse, etc, and it follows my e621-style tags much more accurately whereas all Pony models unfortunately can be pretty random with the results.
Whatever's going on here, it seems like magic, lol. Great work.
I just wish it had the same level of artist comprehension that the aforementioned 1.5 models did.
My only real complaint is that, because it's got Pony in there, human characters swing HEAVILY toward that anime look and it's extremely difficult to get away from that. More comprehensive knowledge of artists would help massively in that regard.
But my previous comment about it being great still stands, lol. :)
It's definitely generating hands and fingers well.
By the way, why the full model weighs 13 gigs and how much can it weigh maximum?
a work this for dragoness?
I'm tempted to assume this model is the FluffyRock of the SDXL Models.
Is this model capable of generating canine and equine pussy, reptile and avian cloaca, arthropod abdomen genitalia, hyaenidae pseudo-penis, equine crotchboobs and bovine and caprine udders?
Does it recognize artist tags?
I did the xyz grid and the echo alpha looks great to me,in comparison to others
why is it 12gb 😭
There seems to be a bit of a problem with this, in that any txt2img sent to img2img will become incredibly dark during the img2img process.