Models developed with a focus on body shape
Please check the images of each version for quality tag samples etc.
v1.2~ optional tag: graycolor_boyish
Please use trigger tags according to your preference.
need: A-Detailer
v1.3 beret mix Manga marge The drawing may be unstable
v2.0 marged base was changed
If the image becomes unintentionally three-dimensional, it may be a good idea to add the sketch tag.
Operation verification environment：https://github.com/Zuntan03/EasyReforge
Description
FAQ
Comments (30)
Now with the new normalize guidance paper lcm models might more interesting
Umm... so like, what's the difference between version 1.5 and other versions? (≧▽≦)
It appears to be a 1.4++ based illustration with enhanced colors.
If the results seem to be similar to other versions, 1.5 may be deprecated just like 1.0 and 1.1 were.
I want to make this model look like a manga, but when I improve the coloring, it starts to look like an anime, which is a bit of a problem.
This is the best model I have ever used!
It's spelled "recommend"
I really love working with this model and use it all the time. I have one question:
Which base architecture does it rely on—SD 1.5 or SDXL? I’m guessing it’s SDXL.
I’m currently trying to apply ControlNet, but I’m having a hard time getting it to work.
情報源
ChatGPT に質問する
This model is based on SDXL (specifically, it's built on illustriousXL), so please make sure to use ControlNet models compatible with SDXL, not the ones for SD 1.5.
GrayColor Thank you very much!!
GrayColor I love this checkpoint very much!
@Takah0x have you tried this one out? I highly recommend it
I didn't get a notification due to a bug on civitai. And I happened to come and see your comment. It's a bit embarrassing (lol). I'll try this model out. :)
This is a very interesting model with unique painting style!
But in my environment, I had a problem with low contrast at the final stage. I tried various things such as quality prompts, triggers, resolutions, and versions (ver1.3 and 2.1), but the same problem occurred. I was able to increase the contrast with an external tool, but when I upscaled with it, the contrast was further reduced, making images look overexposed and difficult to use.(my tools and settings : Fooocus, nvidia video card, eular-a, 30steps)
I think it's because of my tools, but This has never happened with other models. I hope it will be of some help.
I uploaded some samples on my sub account.
Thank you for the detailed feedback — very helpful!
I believe the issue may be due to the fact that this model doesn’t have a VAE baked in. Please try using an SDXL-compatible VAE manually; that should help with the contrast and overexposure problems you're experiencing.
Hope this helps!
GrayColor VAE!! This was a symptom when vae was not available. I didn't notice it because I have not been using a model that required VAE. I tried it with external VAE and it outputted images beautifully. And when used correctly, This model can produce some very interesting and unique images.
Thank you so much for your accurate comment and good model
guys, how to generate colorful pics, i can only generate gray pics now?
The results can be quite random, but I’ve noticed that including the name of the series along with the character’s name as a tag tends to make the background more colorful — or alternatively, using a LoRA can help as well.
I dont know if it helps with every version but if you include the color of items clothing and other stuff then it will be colored in the image. You dont have to name every single color of every single thing.
hands down one of the best models if not the best model around when it comes to aesthetically pleasing body shapes, can use it both for nsfw and sfw images and its working great off the bat without having to also make a soup of tags.
one of my fav models, love the art style and the body shape, hope this model will get updates
very similar to what i wantImage posted by noesca (civitai.com)
v1.3 has issues with high res-fix, image gets washed out + low contrast and there appear little white lights/flashes/spots on picture.
Im downloading v1.4, v1.4++ and v1.5 i hope they will have v1.3 art style without those errors, but it is still great shame that there is such problem with v1.3 :/
I am using sdxl_vae.safetensor
The issue was identified. The VAE on v1.3 is not working properly. After i loaded and used crystalVAESDXL that i had, the washed out colors (grayed out) and white spots - vanished.
Dear creator, those model 1.3 has some secret tag that would force model to do work in color? or bias it strongly/encourage to do work in color? Did it had baked in opposition tag to monochrome/grayscale?
Cause sometimes it's just frustrating, to list out colors for elements, half thru iterations there still is color, and then you look at end results and colors are gone. ofc i could use cfg higier than 1, and benefit from negative prompt, but that will cut speed in half (takes 3 minutes already for 1024) so you understand i'm not willing to call for that solution.
I tried many tags already that should bias model to do colors but they didn't worked. If i have some lora(s preferably 2) enabled this is easy, but without lora (or one lora) this really becomes an issue when i want to achive such color density/vibrance as in your example arts but get black and white pictures instead.
So is there a way to tell model in positive prompt to use colors (or rather keep them, not abandon them half thru iterations)?
Dear creator, those model 1.3 has some secret tag that would force model to do work in color? or bias it strongly/encourage to do work in color? Did it had baked in opposition tag to monochrome/grayscale?
Cause sometimes it's just frustrating, to list out colors for elements, half thru iterations there still is color, and then you look at end results and colors are gone. ofc i could use cfg higier than 1, and benefit from negative prompt, but that will cut speed in half (takes 3 minutes already for 1024) so you understand i'm not willing to call that solution.
I tried many tags already that should bias model to do colors but they didn't worked. If i have some lora(s preferably 2) enabled this is easy, but without lora (or one lora) this really becomes an issue when i want to achive such color density/vibrance as in your example arts but get black and white pictures instead.
So is there a way to tell model in positive prompt to use colors (or rather keep them, not abandon them half thru iterations)?
@GrayColor ? ._.
@AltairTheArc
I made a mistake copying the answer. I am aware that the occasional black and white issue frequently occurs with this model. (It's due to the merged Lora.) However, someone in the community has investigated a method that is likely to improve it, so I'd like to share it. Apparently, adding color to things like hair color, clothing color, and sky color makes it easier to get colorful images.
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.