TroubleDarkness
over 1 year ago

It's 10K downloads my dudes! ๐ŸŽ‰๐ŸŽ‰๐ŸŽ‰

If you ran into any troubles with my mix, please read description below. There is plenty of useful information.

Updates:

  • 02.02.2023 - Added lightweight pruned fp-16 no ema model for DivineEleganceMix and DivinePastelMix.

  • 16.02.2023 - DivineEleganceMix2 released.

  • 17.03.2023 - DivineEleganceMix3 released.

  • 24.03.2023 - DivineEleganceMix4 released.

  • 18.04.2023 - DivineEleganceMix5 released.

Introduction:

All the models i've used before first ever DivineElegance realse did not fit my requirements at all, so i decided to do something by myslef. I saw different mixes built on Anything v4 or v4.5 and i've realized what i truly need. But the main differences between Any | AOM and DivineElegance was the backgrounds.

DivineElegance at the beginning was a mix for my personal use, but it was warmly received by Unstable Diffusion's discord community and some time later i decided to share it, due to many requests.

Right now DivineElegance is my favourite model and quite unique thing for me. The main goal of DivineElegance - anime characters with as many details as it could have. Hair, clothes, backgrounds, all of this things should be beautiful, slightly more realistic and detailed.

I hope you also would like DivineElegace, so stay tuned for newer versions of mix!

Recommendations:

  • Use danbooru tagging for prompts

  • Sampler: Euler at / 30-50 steps

  • Sampler: UniPC at 30 steps for landscapes

  • Cfg scale: 7

  • Clip Skip: 2 <---- VERY IMPORTANT

  • ENSD (Eta noise seed delta) - leave blank

  • Pastel Waifu Diffusion VAE / kl-f8-anime2 VAE

Recommendations (Hires Fix):

  • Upscaler: Fatal Anime

  • Hires steps: 10-15

  • Denoising strength: 0.35

  • Denoising strength (Latent): 0.5

  • Upscale by: any

If you want to get mostly the same results, you difinitely will need negative embeddings:

I'd also recomment it for permanent use.

If you're ran into any problem with my mix, you can dm me on discord: TroubleDarkness#6280, I'll be glad to help | ะ‘ัƒะดัƒ ั€ะฐะด ะฟะพะผะพั‡ัŒ.

DivineEleganceMix V5:

My main model right now.

To merge it by yourself you need:

  • Step 0: Cocoa + AOM3 at Weighted Sum 0.4

  • Step 1: Anything-5 + BasilMix-Fixed

    MBW: 1,0.9,0.7,0.5,0.3,0.1,1,1,1,1,1,1,0,0,0,0,0,0,0,0.1,0.3,0.5,0.7,0.9,1 - Alpha 0

  • Step 2: Previous mix + (NovelAI-NSFW - NovelAI-SFW) at Add Difference 0.3

  • Step 3: Previous Mix + Grapefruit 3.1 at Weighted Sum 0.5

  • Step 4: Previous mix + CocoaAOM3 at Weighted Sum 0.5

  • Step 5: Previous mix + SensuzHolyshit at Weighted Sum 0.2

  • Step 6: Previous Mix + Face-Bomb-Mix at Weighted Sum 0.3

  • Step 7: Convert .ckpt file to fp16 no-ema .safetensors model

Don't save anything before Step 7 as safetensors of half!

DivineEleganceMix V4:

Old version of DivineElegance.

To merge it by yourself you need:

  • Step 0: You need:

  • Step 1: Block Merge (Anything4.5 + BasilMix Fixed) at weights 1,0.9,0.7,0.5,0.3,0.1,1,1,1,1,1,1,0,0,0,0,0,0,0,0.1,0.3,0.5,0.7,0.9,1 and Base Alpha 0

  • Step 2: Previous Mix + NovelAI NSFW + NovelAI SFW at Add Difference 0.3

  • Step 3: Previous Mix + Grapefruit 3.1 at Weighted Sum 0.5

  • Step 4: Previous Mix + Anything V4.5 at Weighted Sum 0.4

  • Step 5: Previous Mix + (Cocoa + AOM3 at WS 03) at WS 0.5

  • Step 6: Previous Mix + FaceBombMix (no vae) at Weighted Sum 0.3

  • Step 7: Convert .ckpt file to fp16 ema-only .safetensors model

Don't save anything before Step 7 as safetensors of half!

DivineEleganceMix V3:

Old version of DivineElegance.

To merge it by yourself you need:

  • Step 0: You need:

  • Step 1: Block Merge (Anything4.5 + BasilMix Fixed) at weights 1,0.9,0.7,0.5,0.3,0.1,1,1,1,1,1,1,0,0,0,0,0,0,0,0.1,0.3,0.5,0.7,0.9,1 and Base Alpha 0

  • Step 2: Previous Mix + NovelAI NSFW + NovelAI SFW at Add Difference 0.3

  • Step 3: Previous Mix + Grapefruit 4.1 at Weighted Sum 0.5

  • Step 4: Previous Mix + Cocoa at Weighted Sum 0.3

  • Step 5: Previous Mix + ZeroTolerances at Weighted Sum 0.15

  • Step 6: Previous Mix + FaceBombMix at Weighted Sum 0.25

Don't save anything above as safetensors of half!

DivineEleganceMix V2:

Old version of DivineElegance.

  • Recipe is lost :(

Actually, the same receipe as V1, but with different weight values and without TensorFix. Receipe would be changed in future releases.

DivineEleganceMix V1:

Old version of DivineElegance.

To merge it by yourself you need:

  • Step 1: Block Merge (Anything V4.5 + BasilMixFixed) at weights 1,0.9,0.7,0.5,0.3,0.1,1,1,1,1,1,1,0,0,0,0,0,0,0,0.1,0.3,0.5,0.7,0.9,1 and Base Alpha 0

  • Step 2: Previous Mix + NovelAI NSFW + NovelAI SFW at Add Difference 0.3

  • Step 3: Previous Mix + AnythingGape (Made of Anything4.5 + Gape60 + NAI NSFW at Add Difference 1.0) + NAI NSFW at Add Difference 0.3

  • Step 4: Previous Mix + Cocoa at Weighted Sum 0.5 = DivineEleganceMix, then, make a TensorFix (aka ClipFix)

That's all you need for base model. Also, you can skip step 3 to avoid more nsfw-ish images in generations.

Really BIG thanks to:

Read more...

What is DivineEleganceMix ?

DivineEleganceMix is a highly specialized Image generation AI Model of type Safetensors / Checkpoint AI Model created by AI community user TroubleDarkness. Derived from the powerful Stable Diffusion (SD 1.5) model, DivineEleganceMix has undergone an extensive fine-tuning process, leveraging the power of a dataset consisting of images generated by other AI models or user-contributed data. This fine-tuning process ensures that DivineEleganceMix is capable of generating images that are highly relevant to the specific use-cases it was designed for, such as anime, landscapes, interiors.

With a rating of 4.97 and over 31 ratings, DivineEleganceMix is a popular choice among users for generating high-quality images from text prompts.

Can I download DivineEleganceMix ?

Yes! You can download the latest version of DivineEleganceMix from here.

How to use DivineEleganceMix ?

To use DivineEleganceMix , download the model checkpoint file and set up an UI for running Stable Diffusion models (for example, AUTOMATIC1111). Then, provide the model with a detailed text prompt to generate an image. Experiment with different prompts and settings to achieve the desired results. If this sounds a bit complicated, check out our initial guide to Stable Diffusion โ€“ it might be of help. And if you really want to dive deep into AI image generation and understand how set up AUTOMATIC1111 to use Safetensors / Checkpoint AI Models like DivineEleganceMix , check out our crash course in AI image generation.

Download (1.94 GB) Download available on desktop only
You'll need to use a program like A1111 to run this โ€“ learn how in our crash course

Popularity

11k 2k

Info

Base model: SD 1.5

Latest version (DivineEleganceMix v10): 1 File

To download these files, please visit this page from a desktop computer.

About this version: DivineEleganceMix v10

Long time no see, comrades, It's been almost 5 months since 9th model version release and now i'm happy to introduce you 10th version of DivineEleganceMix.

New version has came out, so, lets find out what has changed:

Mix:

- Now mix consists of just two models with Anything v5 in base. It was merged in a tricky way, to keep anime face aesthetics, but also keep detailed and realistic-like visual for everything else.

- Updated LoRA, that was merged in model after main mix. Now its dataset has less images, but of all of them are tagged manually not to break any possible concept.

- Mix was designed to work best with "Euler", "Heun" and especially "Euler a" samplers, but of course, will work with any other sampler.

- Mix was designed to work best with EasyNegative (v1) embedding at emphasis of 0.5 - 0.8. Mostly being used for restoring contrast on some backgrounds.

Visuals:

- Adjusted style to keep anime-styled face, but detailed and more realistic-like skin features and hair.

- More details in general, the best way to test it is to generate a forest.

- Less cartoon-ish hair visuals, less "oily" and "plastic" look.

- Adjusted textures UNET block to bring textures back. Textures was completely lost in v7-v8, and v9 still had brush-like stokes insted of most textures.

Characters:

- May require additional prompting because of more precisely tagged dataset images. For example, for squatting with spread legs you need to use "squatting, spread legs" tags combinations, or you'll get just quatting with legs together.

- Improved eyes generation quality, it should be much easier to generate better eyes even with lowres image. Quality still depends on color.

- More anime-styled faces, with bigger eyes and without nose (in most cases). Looks great in my opinion.

- Thinner character outline on generations. In case of simple backgrounds it's just thin, in case of any other background types it's almost unseen or even non-existent.

- Improved generation quality on some clothes, for example - dresses + maid dresses, sweaters, cardigans, school uniforms, white, grey and pink panties. There is a minor issue with striped panties, for some reason it can only be generated in white-blue colors.

- Improved basic nudity concepts further, pussy, cleft of venus, nipples are better again, but i've decided not to keep a focus on sexual acts, because most people still using LoRAs for this kind of stuff. Also, merging NSFW stuff causes a big visuals downgrade, because it is on the same UNET block with skin, faces, hair and background details.

- Bigger nipples and areola for female characters because of a training. There is no need for prompting it. In previous versions nipples and areola were really tiny sometimes.

- On the previous version of the mix, model was overreacting to "curvy", "huge breasts", "plump" and other bodyshape tags. Causing some major issues with nipples, for example. Now it has less effect, but works much better in general.

- Some unnatural hair colors (red, blue, green) may have more realistic accents because of block merging technique. For example, you'll have red hair with brown accent instead of just "flame red". May be fixed in next version of DivineEleganceMix, but i need to exand dataset further.

- Worse finger generation quality, any change in blocks i needed was causing a lot of issues with fingers, but i've managed to keep it as good as possible. Really can't do anything with that, even older merge techniques provides the same result right now.

Lighting:

- Less contrast shadows, images should look brighter.

- Because of lighting change, now it's easier to generate simple backgrounds. For example - white background, simple background, that was pretty hard on previous version of model, because in most cases white backgrounds could be morphed into different shades of gray.

Thanks for keeping an eye on Divine series models and thanks for reading this changes, i hope you'll like new version of DivineElengaceMix!.

15 Versions

๐Ÿ˜ฅ There are no DivineEleganceMix v10 prompts yet!

Go ahead and upload yours!

No results

Your query returned no results โ€“ please try removing some filters or trying a different term.