Please consider Buying Me A Coffee if you like my contributions to the community: https://www.buymeacoffee.com/Rjox
v5.1 - Recommended version. Adds more custom training to v5 while remaining very flexible.
v5 - Flexible with more variety of faces, bodies and settings.
v6 - Less flexible and more of a specific look to the faces and bodies. It's had some custom training done to it, and while it still works well, it was created more for my personal work than for public release. It leans more towards studio backdrops and my custom training.
The RyanBlend models are designed primarily to create photorealistic images of women with a high-end glamour/boudoir photography style, with a very subtle hyper-realistic anime style for facial features and body proportions. It's NSFW friendly, but doesn't have to be used that way.
It's excellent at creating closeup skin texture and high quality faces without using a face restoration model.
I've been a professional photographer since the mid-2000's and specialize in a high-end retouched look for glamour and erotic photography. So I try to make my RyanBlend models close to my own photography and editing style.
Deliberate, phDMix, and Realistic Vision v2
...plus a little:
Analog Madness, HARDblend, URPM, and Perfect World
v5.1 and v6 also have a lot of my own custom appearances for faces and bodies.
poo model
poo
This model is mixed by my own trained models with part of cbimix_v11 and xxmix9realistic_v26, which could generate a lot of beatiful girls. and those style and face should not looks similar as the current hot model, such as Chillout or basilmix.
This model is trained by NSFW contents, but I can't guarantee it can't generate NSFW pictures.
All the preview are generated directly without any addon or extension(e.g. Lora, controlnet...)
This is my first attempt trying to create a Model!
Yet another anime model. Produces a different aesthetic to most other anime models.
I'd appreciate any feedback you can give. NAI VAE is baked, but feel free to use an external VAE, though your results may vary. R2.0 preview images use ClearVAE; this is my recommended VAE for R2.0.
I used easynegative for the preview images.
If you like my work, consider supporting me: https://ko-fi.com/mira6
Favourite influencer
Fia
Hi !
i mix model Newbi Use
I'm sure I'm lacking a lot
Please use it well :D
New Mix Model Name 'V2'
V2 and V3 are completely different mix models.
Visit V3 -> https://civitai.com/models/62602/v3
Models Used in Merge Name
->
masterpiece
best quality
Mixture of multiple loras
Hello, the model was created as an artistic style, the model can do almost anything, the main thing is to follow the promt, hands and eyes looks good for the most cases
Model Information:
This model is generally designed for portraits and full-length anime style photos. Fantastic landscapes are quite decent. And it doesn't require kilometer-long queries to get a high-quality result.
Recommend: DPM++2M Karras, Clip skip 2 Sampler, Steps: 25-35+
This model would not have come out without XpucT's help, which made Deliberate
If you have the desire and means to support future models, here you go:
Advanced Cash - U 1281 8592 6885 , E 8642 3924 9315 , R 1339 7462 2915
PEYEER - P1075963156
I hope you like it, thanks for the feedback
Ether Blu Mix is a stylized anime model specialized in creating character illustrations with vibrant colors and a painterly feel. I consider this my rendition of an AOM3 and pastel-mix style of model.
✨ Please Share Your Cool Creations Below! ✨
Since Ether Blu Mix is based on AOM3 it has strong biases in generating females and suggestive/NSFW content. If you're looking for a model with more flexibility please try Ether Real Mix linked above.
Ether Blu Mix 4 is the stylish evolution of the Blu mix. Ether Blu's base, AOM3, has been rebuilt by adding both of YozoRaAru's beautiful models, "YOZORA" and "Color Box." The result is a model that's capable of creating large images with vibrant colors and details. Though in pursuit of this style, Ether Blu has lost a degree of realism. Hopefully I can experiment and return some of the semi-realistic look, but this current style is closer to what I imagined when I first started rebuilding Ether Blu.
Your prompts might react differently as the base of the model is no longer just AOM3, so please experiment and have fun!
If you prefer the more realistic look of the previous versions, please add "realistic" to the start of your prompts and weight it to taste, e.g. "(realistic:1.3)"
Placing "NSFW" into the negative prompt is recommended if you're concerned about suggestive generations or instances of unprompted nudity. Due to AOM3 this mix is generally biased towards suggestive and NSFW imagery.
Use whatever sampler, steps, cfg you prefer.
Sample images were generated with
Sampler: DPM++ 2M Karras
Steps: 20 - 28
CFG: 7 - 11
Clip Skip: 1
Increasing CFG can lead to more vibrant results.
Negative Prompt Example:
"NSFW, (worst quality, low quality:1.3)"
Waifu Diffusion VAE kl-f8-anime2
Strongly recommended as the model was tuned with this VAE's color and saturation in mind.
Sample images were further upscaled using AUTO1111 Hi-res fix. Please utilize it if you wish to achieve similar results.
Latent (nearest-exact)
Upscale by: 2
Denoising strength: 0.55
Are your results changing too much after Hi-res fix? Please try 4x-foolhardy-Remarci.
4x-foolhardy-Remarci
Upscale by: 2
Denoising strength: 0.45 - 0.55
As Ether Blu 4 leans heavier into the anime aesthetic I recommend using the suggested positive and negative prompts of AOM3.
masterpiece, best quality,
NSFW, (worst quality, low quality:1.3)
The model was mixed and tested using these prefixes and as such will give you the most consistent style results. Of course you're free to prompt however you'd like as booru-style models tend to be quite flexible in prompting.
I personally like to use short prompts mixing prose and danbooru-style tags. Experiment and give the AI some freedom to interpret your ideas!
If you look at the prompts included with the sample images they demonstrate that even with a very loose and vague style of prompting you can achieve fun, high quality images.
General guidelines:
Prompt in the style of anime/danbooru-style models, "masterpiece, best quality, ..... "
More anime style images
Prompt in the style of photography/non-anime models, "professional photograph of ..... "
More realistic images, higher style variety
I have no training or real knowledge of machine learning. I merged the model using the information provided on WarriorMama777's AOM model page. I very arbitrarily used the UBM weights given to merge and create AOM3 and its derivatives and modified the values with trial and error until I was visually happy with it.
I utilized the SuperMerger extension for AUTO1111 by hako-mikan.
▼ Merging
MODEL A: AOM3.safetensors
MODEL B: nightSkyYOZORAStyle_yozoraV1PurnedFp16
MODEL C: AOM3.safetensor
Add Difference merge - Don't Use MBW
Alpha: 0.5
Model: AOM3-YozoraDiff-05
▼
MODEL A: AOM3-YozoraDiff-05
MODEL B: colorBoxModel_colorBOX.safetensors
MODEL C: AOM3.safetensors
Add Difference merge - Don't Use MBW
Alpha: 0.3
Model: AOM3-YozoraDiff-05-ColorBoxDiff-03
▼
MODEL A: AOM3-YozoraDiff-05-ColorBoxDiff-03
MODEL B: Counterfeit-V3.0_fp16.safetensors
MODEL C: pastelmix-better-vae-fp16.safetensors
Weight Sum merge - Use MBW
Alpha: 1
0,0.5,0.5,0.5,0.5,0.5,0,0,0,0,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0.5,0,0
Beta: 1
0,0.3,0.3,0.3,0.3,0.3,0,0,0,0,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0.3,0,0
Model: AOM3-YozoraDiff-05-ColorBoxDiff-03-Counter3SumS2-05-PastelSumS2-03
▼
MODEL A: AOM3-YozoraDiff-05-ColorBoxDiff-03-Counter3SumS2-05-PastelSumS2-03
MODEL B: ayonimix_V5.safetensors
Weight Sum merge - Use MBW
Alpha: 1
0,0.4,0.4,0.4,0.4,0.4,0,0,0,0,0.4,0.1,0.4,0.4,0.4,0.4,0.4,0.3,0.1,0.1,0.4,0.4,0.2,0.4,0.4,0.4
Final Model: EtherBluMix-4
Models Used:
AOM3.safetensors
D124FC18F0232D7F0A2A70358CDB1288AF9E1EE8596200F50F0936BE59514F6D
Counterfeit-V3.0_fp16.safetensors
CBFBA64E662370F59D4AA2AA69BF16749FCE93846CCCE20506AEE5DF01169859
pastelmix-better-vae-fp16.safetensors
D01A68AE76F97506363F387F5F28BB564AD9E20924844FD5945E600B72D39E79
colorBoxModel_colorBOX.safetensors
93A20525F51798309AF22AC6E8285E61CF563A88F1EE00AA9F544CA939E63181
nightSkyYOZORAStyle_yozoraV1PurnedFp16
E7BF829CFFD9EA034DA3B59FA13421F52AA4E4ED74FCBBEEFBF9CFB2F97C329E
ayonimix_V5.safetensors
B46B04432B070FD88D69161B62C43EDCD4340DFC27FFB2E5FE7EDA64950C7E34
Thank you to WarriorMama777 for providing AOM and other various mixes as well as detailing your workflow and inspiring me to try mixing my own models.
Thank you to hako-mikan for creating the SuperMerger extension allowing a quicker workflow for merging.
Thank you to gsdf, rqdwdw for creating Counterfeit.
Thank you to YozoRaAru for Night Sky YOZORA and Color Box.
Thank you to andite for Pastel-Mix.
Thank you to Ayoni for AyoniMix.
Thank you to AUTO1111 for creating the web UI everyone uses.
Thank you to StabilityAI for starting everything with Stable Diffusion.
Thank you to the entire SD community for continuing to openly share and create.
GirlsMix (koreangirlmix)
Used to create a beautiful girl with a variety of traditional costumes.
Limit errors in hands and eyes, there are still errors but the rate is low.
realistic
Modelo personal 1.5 - The keyword to create the character is makedata
The eighth merge I've made, and the first one I'm rather happy with.
generates unrealisitc bodies with semi-realistic faces.
Uses vae-ft-mse-840000-ema-pruned baked in.
Recommend using clip skip 2.
用来自己看看的
帅哥
I made a disclamer, so download it and use only if you struggle with my frames Loras (or any other, that use 2x2frames or 3x3frames tags). In short - it's a model that knows a basic concept of frames as a composition class, if that makes sense. Supposedly it gives more control of grids Loras, and at the first glimpse it really does.
This model was made like this:
Firstly I trained a model on about 500 images of frames;
Then I clean it from any traces of the model I trained it on through merging;
Then I merged it with perfectDeliberate model a couple of times to get this result.
Basicaly it's a perfectDeliberate model with frames, but I found a couple of downgrades in quality on my model when generated casual (not frames) pictures. So if you want to use as a standalone model - just use perfectDeliberate instead, it's just better.
Download any of the grids Loras;
Use this model as "guide" in txt2img (just generate a couple of pictures in 512x512 (2x2frames tag) or 768x768 (3x3frames tag), till you are satisfied with the results, there will be artifacts, because images are small);
Then change the model to your liking, set denoising str 0.5-0.6 and lower the Lora weight to 0.4-0.6, and upscale it as much as you can;
Also you can make gifs whith this model alone, the best results I get was with hiers fix on 0.5 str.
Guide on how to use my grid Loras.
Sometimes here and there appears strange "glowing" artifacts. I guess it's because of the baked in perfectDeliberate VAE, or maybe I did something wrong when merging models. I'm trying to get rid of it.
Feel free to ask any questions here in comments, or in my discord channel.
Also I am making games with AI arts. They will be free in the future, but if you want to participate in making or have an early access - you can support me on patreon)
Also my wet hair Lora was banned here, I dunno why. You can download the last version of it from my patreon for free.
2x2frames
3x3frames
It has good detail and physical expression
nai's vae is enough
High variety anime model ~ ~
🐐🐐🐐🐐🐐🐐🐐🐐🐐🐐
////////////////////////////////////////////////////////////////////////////////////
Recommended Settings
Clip skip : 2
Hires. fix : R-ESRGAN 4x+ Ultrasharp / Upscale 1.8 + / Hires steps 8 / Denoising strength: 0.5
Sampler : DPM++ 2M Karras / DPM++ SDE Karras / DDIM
CFG : 7 Steps : 20
Prompts
Original Character, Volumetric Lighting, Best Shadows, masterpiece:1.3, best quality
Neg : (Worst Quality, Low Quality:1.4), Poorly Made Bad 3D, Lousy Bad Realistic
////////////////////////////////////////////////////////////////////////////////////
If you enjoy my art consider checking out my other models!
Moe needs ko-fi too 😿😻😻
////////////////////////////////////////////////////////////////////////////////////
By downloading this model you agree to assume full responsability on how you use it! Author (me) has no liability.
Takeda Hiromitsu style, masterpiece, highest quality, official art, nsfw, 1girl, girl room, bink bed, 1girl, (amaya haruko:1.0), dark blue hair, blue eyes, (gradient pink at the tip), (pink sweater:1.0),( blue short jeans:1.0), backless outfit, long ponytail, aroused, seductive smile, smirk, looking at viewer, large breasts, sexually suggestive, sexual hand gesture, depth of field, bloom, dynamic lighting, skindentation, skin tight,
girl, amaya haruko, dark blue hair, blue eyes, (gradient pink at the tip), (pink sweater:1.0),( blue short jeans:1.0), backless outfit, long ponytail, aroused, seductive smile, smirk, looking at viewer, large breasts, sexually suggestive, sexual hand gesture, depth of field, bloom, dynamic lighting, skindentation, skin tight, Takeda Hiromitsu style
excellent for making futanaris
futanari
这个ckpt模型是我1月份左右融合的(或者更早),最近在测试训练好的lora的时候翻到了,才想起来有这个东西。这个模型是PVC模型和一个真人模型融合而成的,具体是哪一个我已经忘了,但是没猜错应该是basil mix。融合出来的表现就是很像充气娃娃,感觉还挺好玩的。
The CKPT model is a merge ckpt I made around January(maybe more early). I came across it while testing the well-trained Lora recently, and then I remembered that such a thing existed. This model is a fusion of PVC model and a real human model, but I've forgotten which one it was specifically. Nonetheless, I believe it is likely the "basil mix". The resulting appearance is quite similar to that of an inflatable doll, which was rather amusing.Not a special VAE is required.
模型需要加载vae,但是不需要特别的vae,用自己经常用的就行
The model requires loading VAE, but a special VAE is not needed. You can use the one you usually use.
这个模型使用了多个lora模型进行合并 使用的底模是 AnythingV5_v5RE.ckpt
This model uses multiple LoRa models to combine, and the base mold used is AnythingV5_v5RE.ckpt
使用这个模型 配合lora和高清修复,可以做到完美的瞳孔与细节
With this model with LoRa and HD Restoration, you can achieve perfect pupils and details
1girl
cat_ears
A model for creating photorealistic images.
This version used BLIP2 captioning, and a mixture of nude and non-nude images with cinematic images in the mix. The following is different from V2:
Improves full body creations while sacrificing face fidelity. Use it with face restore.
Produces less NSFW compared to V2 unless prompted otherwise.
More mature faces.
Generates cinematic images.
Works well with LoRAs.
Can be used with higher CFG.
Sample prompts:
"a young woman, street, laughing, ponytails, (hdr:1.3), (muted colors:1.2), dramatic, complex background, cinematic, filmic, (artstation:0.8), soaking wet"
"photograph of a girl, street, <lora:A-Mecha-REN[color_theme,mecha musume, mechanical parts, robot joints,headgear,full armor]:1>,black_theme,mecha musume, mechanical parts, robot joints,headgear,full armor,ponytails, (hdr:1.3), dramatic, complex background, cinematic, filmic, <lora:GlowingRunesAIV6:0.8>, GlowingRunesAIV2_red"
"photograph of a woman, street, skirtlift, shirtlift,<lora:skirtlift-v1:1>, (hdr:1.3), (muted colors:1.2), dramatic, complex background, cinematic, filmic, (artstation:0.8), soaking wet, <lora:shirtliftv1:1>"
Negative prompt:
"child, deformed, deformed face, unreal, bad hands, bad fingers, monochrome, over saturated, bad hands, lowers, 3d render, cinematic, cartoon, long body, blurry, duplicate, duplicate body parts, disfigured, poorly drawn, extra limbs, fused fingers, extra fingers, twisted, malformed hands, mutated hands and fingers, contorted, conjoined, missing limbs, logo, signature, text, words, low res, boring, mutated, artifacts, bad art, gross, ugly, poor quality, (low quality:1.2), (worst quality:2)"
"(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, tattoo, extra heads, art, unreal"
Scale:
5-10
Steps:
20 - 50
clip_skip: 2
Very NSFW even with negative prompt added keywords "nude" and "naked", but can also create SFW.
Sample prompts:
"close up of a black girl smiling looking at the camera"
"highly detailed photography, spring, portrait of beautiful blonde woman standing in a park, top and miniskirt, perfect face, perfect body, large breasts, slightly thin skin, sharp focus, photo by Annie Leibovitz, intricate, natural lighting, high quality, 4k, cover photo"
"a 25 year old European girl with red hair wearing a white bikini top and denim skirt, posing for the camera by the beach, in a boat"
Negative prompt :
"nude, naked, child, deformed, deformed face, unreal, bad hands, bad fingers, monochrome, oversaturated, bad hands, lowers, 3d render, cinematic, cartoon, long body, blurry, duplicate, duplicate body parts, disfigured, poorly drawn, extra limbs, fused fingers, extra fingers, twisted, malformed hands, mutated hands and fingers, contorted, conjoined, missing limbs, logo, signature, text, words, low res, boring, mutated, artifacts, bad art, gross, ugly, poor quality, low quality"
Scale:
3-5
Steps:
50 - 100
clip_skip: 2-4
This model is a mixture of the models that I have liked the most after trying many and having tried many recipes. Can output realistic images of both sfw and nsfw.
Used embeddings: bad_prompt [1d99], bad-artist [2a38], bad_prompt_version2 [afea], easynegative [119b], EasyNegativeV2 [5685], bad-picture-chill-75v [1eff]
A new blend I'm testing. I use it often for lora testing so I thought I would upload it.
Body Type: Double Thick Thighs, Perky Breasts, Round Breasts, Max Cleavage, Amazon Goddess,
Fashion: Crop Tops, bodycon minidress, Scrunchbutt leggings, Cybergoth, Nerdy, bodycon skirts, bodycon shorts
Environment: Undersea Depths
This is my second model merge. It uses 8 different models including a custom tuned SD1.5 base. Please share and enjoy.
後半に日本語の説明を入れています。
This is a 2D based Anime model (not using 2.5D approach). Targetting to realize both simplified and clear drawing and enough detail.
There are 3 variation for this model. Base model (v1.15), more outline version (v1 Line) and less outline version (v1 Flat). You can switch among them to optimize your output. No variation from ver3. Use outline LoRAs (plus or minus) for control outline.
Merged models are as follows.
Counterfeit v3
Anything v5
Falkons v1.2 created by Falcons
Contaby v2 created by PlayaEmblem
Koji
LinexDiffusion / ALLBoyMix (after version 3)
Both SFW and soft NSFW are capable. VAE is not needed but you can try some.
2Dベースのアニメ系モデルで、いわゆるリアルモデルによる2.5Dのアプローチを排除したモデルです。シンプルではっきりした塗りと細部の詳細な表現の両立を目指したモデルとなっています。
3つのバリエーションが用意されています。ベースモデル (v1.15)、輪郭線を強調したバージョン (v1 Line)、輪郭線を削ったバージョン (v1 Flat)です。好みに合わせて使い分けてください。 ver3でバリエーションをなくしました。アウトライン系LoRA(プラス/マイナス適用)で調整してください。
Counterfeit v3、Anything v5、Falkons v1.2(Falcons氏作)、Contaby v2(PlayaEmblem氏作)、Koji、LinexDiffusion(ver3以降)、ALLBoyMix(ver3以降)がマージベースです。
微エロおよび非エロに対応可。VAEは焼き込み済ですが別のものも試してみていいと思います。
This model is the fifth generation of the Defacta family.
It has more advanced background rendering and fewer broken characters.
EasyNegativeV2 recommended
Expmix variation model.
This model show 2D anime style using Flat colors and strong outlines
V3 update
Show cleaner line and more vivid color.
Also show softer skin texture and lighting than V2
Expmix_line_2
Treebark
Mochizuki Kei (望月けい) Art Style LoRA
Expmix_Line_V2 + Nabimix (Block merge)
0,0,0,0.15,0.3,0.1,0.1,0.2,0,0,0,0.2,0.3,0.2,0,0,0,0.25,0.2,0.3,0,0,0,0,0 = A
A + Counterfeit V3 (Block merge)
0.25,0.5,0.15,0.3,0.1,0.2,0.2,0.15,0.3,0.2,0.4,0,0.3,0.1,0.1,0,0.2,0,0,0,0,0,0,0.2,0.2 =B
B + 0.2(Mochizuki Kei Art Style LoAR) = C
C+Treebark (Block merge)
0,0,0.1,0.1,0,0,0,0.05,0.1,0.1,0.1,0.1,0.1,0,0,0,0,0,0,0,0,0,0,0,0 = expmix_line_v3
Sampling method : DPM++ SDE Karras, DPM++ 2M Karras
Clip skip : 1~2
Hires steps : 13
Hires.fix upscaler : R-ESRGAN 4x+Anime6B, DDIM
CFG Scale : 7~10
VAE : BerrysMix = (kl-f8-anime2)
Prompt : (masterpiece:1.2, best quality)
Negative : (worst quality, low quality:1.4)
Textual Inversion : negative_hand-neg, verybadimagenegative
A new blend I'm testing. I use it often for lora testing so I thought I would upload it.
Body Type: Double Thick Thighs, Perky Breasts, Round Breasts, Max Cleavage, Amazon Goddess,
Fashion: Crop Tops, bodycon minidress, Scrunchbutt leggings, Cybergoth, Nerdy, bodycon skirts, bodycon shorts
Environment: Undersea Depths
Pos: masterpiece, best quality
neg: (worst quality, low quality:1.4)
Done.
I prefer DPM ++ 2M Karras as main sampler, seem to work best in most cases. 12-20 steps.
Hires fix - anything you like. Lowest denoise boundary seem to be 0.53.
Yet another anime model mix!
Purpose of this model is to be decently diverse, high quality(obviously...), editable and not very leaky, while maintaining high usability and benefits of underlying models.
Or so i would say, if that was the truth.
Real story of this model is that i explored deep merge a bit back in the day, because i was inspired a bit by exploration of block merged by AOM2. So, here i go down the rabbit hole... This model is a result of my nightly exploration in the dwellings of by-block merge.
I can't say this is a very precise one, but it is also not a random guess. I loosely evaluated changes of IN00-IN11 blocks, and proceeded with merging all of blocks with rough estimation of values i would prefer. That was step one. This mix is double-deep-merge, and second step was mostly a shot in the dark, based on info i already got. It worked out first try, believe it or not.
I honestly don't remember exact values, i remember i saved them somewhere, and i will add them later if i would find them.
First step was merging "ACertainModel" or "Anything3"(or any other base model of your liking with similar style) with values that roughly add up to 43/57, but vary highly from block to block.
Next step is to sprinkle specific blocks of AOM2 in to that mix.
Though, remember, mixing mixes is bad, it's hard to train, especially with D-Adaptation. But it eats all kinds of loras well. Preferrably trained on NAI.
ACMAR G Mix is general anime model, so it will be good at pretty much anything that you would expect from general anime model, including utilisation of LoRAs(this is what many more stands for)
It has enchanced capabilities at 2.5 and 3D thanks to AniReal and AOM2 included in mix.
It is unlikely to be capable of hard nsfw without LoRAs, but general one is, of course, very basic feature.
Lineart/sketch style is quite decent too.
Lower bound for denoising in hires fix seem to be 0.53, which is quite low.
This model is the one im using to generate most arts for previews of my LoRAs, so you will be able to repeat them.
For some reason it's capable of generating 1024x1024 base resolution decently and reliably enough, at least in my case. Couple times i even got it to generate good 1536x1536 natively(no hires. fix).
(This is rare 1536x1536)
(And this is common 1024x1024 hires to 2048x2048)
It is usually more stable than AOM2 at more unusual aspect ratios and composition in general, at least from my tests, but be aware that it can be just random, as i didn't test extensively.
Also btw, don't you also love extra wide face shots?
Extra long (3:1(512x1536)) generations are possible natively too, quite stable. Same goes for wide.
It is not very vital to train based on this model, other than embeddings. It will overtrain very easily. If you are a user of D-Adaptation - don't bother, i spent days figuring out various parameters and still they weren't behaving as expected at higher ranks...
Also my internet is too bad to upload FP32 initially, so you'll have to wait :D
Hey, if you have read all that, i don't like to ask that, but i would really appreciate a monetary support - https://www.patreon.com/anzhc
This will help me with training and providing AI stuff further.
I will be working with this model more.
The progress I've been making this is pretty nice, if there any suggestion guys of what other model's should I merge it with?
Yuzus goal are easy to archive high quality images with a style that can range from anime to light semi realistic (where semi realistic is the default style). Be aware that some prompts can push it more to realism like "detailed". If faces apear more near the viewer, it also tends to go more realistic.
My Discord, for everything related to anime models. Baked in VAE is clearvae_variant. You can also use klF8Anime2, to push the colors, but eyes can get blurry.
Sponsors:
AnimeMaker.ai is a proud sponsor of hassaku, preinstalled with 300 LoRAs & 100+ full NSFW models. Render fast from ANY device. Join their community at https://animemaker.ai/
Mage.space with its amazing creators program supports all kinds of creators like me! Preinstalled with 50+ high quality models, join their discord community here!
SinkIn.ai hosts the best Stable Diffusion models on fast GPUs. You can run Hassaku at: https://sinkin.ai/m/76EmEaz
Supporters:
Thanks to my supporters Riyu on my patreon!
You can support me on my patreon, where you can get early access to newer model versions.
_____________________________________________________
Using the model:
Use mostly danbooru tags. No extra vae needed. For better promting on it, use this LINK or LINK. But instead of {}, use (), stable-diffusion-webui use (). Use "masterpiece" and "best quality" in positive, "worst quality" and "low quality" in negative.
My negative ones are: (low quality, worst quality:1.4) with extra monochrome, signature, text or logo when needed.
Use clip skip 1 or 2 with sampler DPM++ 2M Karras or DDIM. I use clip 2.
Don't use face restore and underscores _, type red eyes and not red_eyes.
Version 3.1 is a hotfix, as the prune for v3 failed and was showing errors on peoples pictures.
(It is very dependant on negative prompts, so use the example images as a base.) The first image uses a light map as img2img so your results may very, but i thought it looked cool so i included it.
The model is better with a VAE. (vae-ft-mse-840000-ema-pruned is fine or you can use your own.)
* For some reason (for V2) the pruned version isn't showing as the first download and the hashes have got screwy. Please use the dropdown menu to choose the smaller version if desired *
This model is a custom blend of various models, (see below), giving many options for generating a wide variety of images, including NSFW. Based on Stable Diffusion 1.5.
Primarily focused on capturing skin texture and the feel of photography, and Polaroids in general, but experiment with other things. (This model is not perfect).
Though there are is no specific KEYWORD to 'turn it on', if you use words in the prompt such as 'polaroid' , 'analog' , 'film grain', 'perfect eyes' and so forth at the start of the prompt, you should get the look you're after.
The example pictures do not have any face/eye correction or upscaling. 768 x 1024px is quite stable at producing images without 'mutation' if the prompt and neg prompt are good.
The model needs a VAE. (vae-ft-mse-840000-ema-pruned is fine or you can use your own.)
Prompt example:
(polaroid), portrait, ((extreme realistic skin detail)), (20yo) (SUBJECT NAME or PERSON TYPE), ((holding a heart balloon)), lingerie, sassy, short black hair, sharp focus, detailed skin, detailed face, skin pores, skin imperfections, film grain, dynamic lighting,
Negative:
Child, jewellery, Plastic skin, Overexposed, (semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, digital art, anime, painting, manga), (deformed iris), distorted body, (deformed pupils), grit, man, distorted, (extra arms), long body, male, skin rash, deformed body, mutated hands, poorly drawn eyes, head crop, smooth skin , deformed hands, (deformed proportions) , long face, long neck, looking away, puffy nipples, watermark, logo, text,
or something like:
child, blur, old face, extra arms, big chin, (long neck!), ((more than one face)), imperfect eyes, glowing eyes, (head crop), pout, (abnormal body), double head, oversaturated, (head out of frame), too close, bad hands, distant, ugly, male, fake breasts, tiling, large nipples, extra heads, deformed, cgi, anime, painting, boring pose, plastic skin, extra limbs, big forehead, abnormal, long body, long limbs, overexposed, text, watermark, hairy body
A non-exhaustive lost of models merged (because I didn't take notes): deliberate, clarity, dreamlike-photoreal, realistic vision, protogen (+ additional ones in version notes).
Made for fun, not for money or notoriety. Those who train the models from scratch are the ones deserving of applause.
Have fun! Can't wait to see what you come up with...
2.5D- The final version of SAV_MIX, tweaked for anime with the addition of a large amount of 2.5D. Even more details and naked bodies. No need for nudity? You know what to write in the negative, the details don't suffer from it.
Somehow I managed to achieve a result in the combination of my favorite genres: fantasy and anime. Judge for yourself, share the results!
There should be many beautiful words why this mix is better than others, but there won't be any, the best it is only for me. There are a dozen LORA models assembled, anime stylization (I'll be honest with the hentai), one realistic model - heavily influenced this mix and at the end I added an emphasis on landscapes. The landscapes, even with minimal querying, turn out surprisingly well. And yes, don't be surprised if it suddenly gives you monsters.
I recommend using it in conjunction with: vae-ft-mse-840000-ema-pruned
Good day, everyone!
Thank you for your interest in my model, this model is based on my previous model "Pika's Animated Mix"
But what's new?
I trained this model on a style which is very close to the style of Niji V5, it has a strong cinematic effect and requires very grammatic input prompt to get quality images!
Also the model is very flexible in style, try it yourself!
The model comes without VAE!
Here are my recommendations:
Sampler: Euler a, DPM2 a Karras, UniPC
Steps: 28-50+
CFG: 4.5-8
VAE: Any Anime VAE you like.
Clip Skip: 1,2
Negative prompt: (easynegative:1.1), (verybadimagenegative_v1.3:1), (low quality:1.2), (worst quality:1.2)
Enjoy using it and be sure to share your results!
P.S: I hope your results are better than mine! :D
Looking forward to your feedback <3
This pet project is my custom (mostly NSFW) photorealistic model, blended and fine-tuned on hand-captioned images of people, both clothed and nude, mostly from professional photo shoots. My goal is to create a model that generates beautiful images of people by default without needing excessive prompts, so that your prompts can focus on adding qualities instead of removing defects.
I encourage you to try it without too many negative prompts, embeddings, or textual inversions first and only add those in as needed. You also may need to reduce the weight of LoRAs that impose similar effects as the model itself.
Real Hot Mix - a model that looks like real life; no uber contrast & oversaturated colors, no fake plastic skin when generated or upscaled, just normal real life images with:
realistic skin textures!!!!!!!
(dimples, bumps, freckles, imperfections)
realistic lighting
flexible prompting (poses, anatomy, backgrounds)
easy to prompt for large OR small breasts
This model is a mix, so no new content was trained. I just wanted a model that was easy to work with and a good base to train future LoRAs on. Enjoy!
(All images & models are prompted with 18+ descriptions & trained with 18+ models!)
Mixing (what I believe to be) the best qualities from several photo-real models:
Without getting into too much detail, I think LazyMix is one of the best to prompt with (especially for NSFW women), but has a plastic/flat look when upscaling and going for a photorealistic look (vs. analog/grainy/selfie quality). Meanwhile, Edge of Realism and Realistic Vision are very photoreal, and Analog Madness is sort of a hybrid. My intention was to blend LazyMix and Edge of Realism as the primary models, and then add flavors of Realistic and Analog. Real Hot Mix should capture prompt flexibility, NSFW bits, lighting, and skin textures all in 1 model.
Mix Formula:
0.9(0.85(0.5(lazymixRealAmateur_v20) + 0.5(edgeOfRealism_eorV20Fp16BakedVAE)) + 0.15(realisticVisionV20_v20)) + 0.1(analogMadness_v40)
Respective Weights:
0.3825 LazyMix+
0.3825 Edge of Realism
0.1350 Realistic Vision
0.1000 Analog Madness
I find that generating txt2img (with ControlNet 'softedge' or 'openpose' if desired) with height/width <=768px and then upscaling (hires fix or img2img) with Nickelback (or maybe NMKD) upscalers works well
For upscaling, check out this vid by Sebastian Kamph
For inpainting, set masked content = original
inpaint initial txt2img at either whole picture or only masked
inpaint upscaled image with only masked
large padding - idk exactly, but I usually like something larger than the default
modify prompt as needed depending on what is/isn't masked
helps if you know photoshop, edit, then reupload to img2img inpainting
can draw in parts with brush tool
can fix parts with content aware/liquify (ex. iris)
experiment with ControlNet
Contact me at aihotgirls@proton.me
Mix between chilloutmix and perfectWorld with a bit of koreandolllikeness
I'm creating some amazing art on my Patreon and accepting commissions:
Follow all my work on my Instagram:
This model was put together to make anime like stuff
Clip = 1
Queencomplex1