🟡:Flux Models 🔵:SD XL Models 🔴:Expired Models
🟡 STOIQO Afrodite is a specialized model designed to generate high-quality adult photography content. It allows users to create realistic and explicit images with a strong emphasis on fine details and authentic textures. Perfect for those seeking precision in the adult photography space, it provides creators with the ultimate tool to produce high-end explicit visual content.
Versions and Recommended Settings:
MAIN SAMPLER: Euler+ Beta
GUIDANCE (In Forge:'Distilled CFG Scale'): 2+ | STEPS: 15+
(Recommended): GUIDANCE: 3.5 | STEPS: 25
We finally got to the release of the flux.1 Dev version of Afrodite. As for NewReality we are still in an experimental Alpha version, which still has many problems, but which I will be able to intervene thanks to your feedback.
The Main Download (Full 11GB) is the version containing only the Unet, so you can use it accompanied by the clips of your choice, depending on whether you prefer the fp8, the fp16 or want to experiment with new clip_l or t5.
In the file section of the sidebar you can instead find the Secondary Download (Pruned 20GB) of the AIO version with t5xxl_fp16, clip_l and VAE already included.
The Full/Pruned nomenclature is just to be able to keep the Unet as the main download. All the others change with the AIO as the main. The two models are composed as mentioned above.
Refiner is unnecessary and Clip and VAE are included
MAIN SAMPLER: dpmpp_3m_sde + exponential
CFG: 2+ | STEPS: 16+
(Recommended): CFG: 4 | STEPS: 25
Credits:
Afrodite is the result of countless hours of work, involving several hundred iterations through training, merging of models and LoRAs, fine-tuning specific blocks or captions, as well as performing additive and subtractive merges. Due to the complexity of this process, it is difficult to provide a precise, step-by-step account of every decision and experiment. However, each phase has played a significant role in shaping the final product, whether by direct influence or through valuable lessons learned along the way.
Given the intricacies and challenges involved, I want to extend my heartfelt gratitude to the creators of the models that have contributed to this journey, either directly or indirectly. Regardless of whether they were incorporated into the final iteration, their work provided inspiration, insight, and progress throughout the project.
To all the creators whose models supported this endeavor, I extend my deepest thanks. I encourage everyone to explore and support their work, as it deserves recognition for the incredible value it brings to the community.
EauDeNoire humblemikey socalguitarist ZyloO xlabs_ai aramintastudio Seeker70 Ai_Art_Vision PromptoAI alvdansen aip0pp SG_161222
WORK IN PROGRESS...
Description
FAQ
Comments (118)
The 20GB version is not in the download section.
Will it be added in later?
what is the point though? unless the clip is trained its just bloat
@jensgruber8659 I had an error while uploading the model, it is uploading right now. Please try again later to see if it is available.
@low_rank sorry but i don't understand what you mean. True that working also on the clip you have an additional boost on the generation, but judging by the results, it doesn't seem to me to be just a bloat. What are the things that don't convince you? do you have examples? how can i intervene to improve future versions?
Thanks for not asking for Buzz like many others do nowadays 🙏
I agree but I don't mind people asking for a reasonable amount of buzz for early access, as long as the model looks to be high quality and the description and images are well documented. That's a lot of work and deserves some compensation. I think 50-200 buzz is a fair amount for early access in those cases.
The ones that ask for 2000-5000 buzz are going too far, especially if they have almost no description and just a few crappy images that don't look any better than what you'd get from the base Dev model.
hello @svooooosh , I started this project almost for fun with the idea of providing access, first of all to myself, to an uncensored and free model. Even if the numbers and visibility have increased I would like to continue to maintain these two characteristics in my models. Earning a few more Buzzes does not make the difference to me as much as feedback and appreciation from you do. I am evaluating methods to monetize, but don't worry it will not be on the cost of purchasing the models.
Nothing negative towards those who offer the paid model, provided that as @Grumblebutt also says, it is not out of scale with the offer and that the informations about them are clear.
pity no male genital exposed
Yeah let's see some weiners.
@itasky @girlswithafros I'm working on that too. Flux is much better than XL, but the successful versions with male anatomy didn't satisfy many other parameters. The female dataset is still too large and too captioned. I hope to be able to provide a more or less working model with male nudity in the next version.
For now use some Loras, and indeed if you test them with my model also post some feedback on the compatibility that always helps to improve the finetuning.
Please train weiners Bro.
Looking forward to trying this one later this weekend. I have my GPU working on a DEV finetune now, so have to wait a little longer. Glad to see some trained models having success though. Still trying to figure out the best raining settings.
hi, I am confused about the labels of F1D Alpha versions.. the pruned is larger than the full which is unusual. By the way... are these unets or, have all the clips and vae added?
Just have to read the readme of the model.
I quote:
The Main Download (Full 11GB) is the version containing only the Unet, so you can use it accompanied by the clips of your choice, depending on whether you prefer the fp8, the fp16 or want to experiment with new clip_l or t5.
In the file section of the sidebar you can instead find the Secondary Download (Pruned 20GB) of the AIO version with t5xxl_fp16, clip_l and VAE already included.
The Full/Pruned nomenclature is just to be able to keep the Unet as the main download. All the others change with the AIO as the main. The two models are composed as mentioned above.
what u have merged ?
Beneficiaries of other work ... do not mention what all users doo for you
Any way we could get quantized versions?
I made Q4_0, Q5_0 and Q8_0 quants but the Q8_0 one is to big for me to upload so you have to work with Q4_0 and Q5_0 for now.
If @ALIENHAZE doesn't want me to distribute the quants like this please tell me and I will delete the comment and the files from google drive!
https://drive.google.com/drive/folders/1h2zW3TDO7r6NQVre5xYd6qNAw7V2qyua?usp=sharing
@firedelta4 Can you reupload please?
@KenHik The link still works for me. It links to a folder with the Q4_0 and Q5_0 gguf files. tested it in incognito mode so I was also not logged in. What error are you seeing?
@firedelta4download quota exceeded. sorry, I don't use google drive. How to download it?
@KenHik ok thanks for the heads up. The files might have hit some abuse protection or whatever bcs they were downloaded a lot and they are big files. will search for alternative means of hosting although that is hard with big files
@KenHik I don't know how fast the download speeds are after you hit the 6GB quota but this is the best thing I could come up with in a short time.
https://pixeldrain.com/l/HuJjgvcU
@firedelta4 Thank you very much! Pixeldrain is great, one of my favorite. It's 1 MB after 6GB.
Why CPU usage so high with this model when running in ComfyUI?
This looks great! I hope you can make a GGUF Q4 or Q5 so more people can use it
I made Q4_0, Q5_0 and Q8_0 quants but the Q8_0 one is too big for me to upload so you have to work with Q4_0 and Q5_0 for now.
If @ALIENHAZE doesn't want me to distribute the quants like this please tell me and I will delete the comment and the files from google drive!
https://drive.google.com/drive/folders/1h2zW3TDO7r6NQVre5xYd6qNAw7V2qyua?usp=sharing
Alternative download source if google drive might give you trouble:
https://pixeldrain.com/l/HuJjgvcU
@firedelta4 The Q4 keeps giving an error "mat1 and mat2 shapes cannot be multiplied" when i try to generate anything. Any idea why?
@narugoku what UI are you using?
I only tested in Comfy UI with the Comfy-GGUF nodes. I don't know about the support for gguf in forge or other UIs.
If you use comfy please show me your workflow
@firedelta4 Thanks for this it works grate on invoke.
If you could please provide optimal settings for Steps Sampling+Schedule, Distilled CFG scale + CFG scale.
Thank you : )
Great model, will be sending some buzz your way as a form of thanks, by the way for your flux models, what are the key prompts used for skinny flatchested woman? thnaks!
Works just as good as your compact Flux models.
4060 Ti 16gb has no issues with Pruned F1.D Alpha fp16 20gb model at 2.3s/iteration.
This model changed my view on Flux.1 for the better. I hope you bring more of these merges!
Thank you!
Not bad, but also not unique. Yet another Flux.1 dev that can produce a small number of nude women in a small number of poses, both of which very strongly influence anything tangentially related to it. See the nipples fused into the ladies' clothing in the pics below? It's because the finetuning data is too heavily focused on nudity. Even if nudity is the goal, there needs to be a balance when introducing "novel" concepts to the base model (as in, literally any adjective next to the word "breast") or it will start to neglect what clothing actually looks like, thus harming its ability to image cloth that isn't see-thru, or poked through by superpowered nipples in January.
Can't solve this with merges, btw, gotta roll up your sleeves and tune. Happy hunting~
How to train Lora on this checkpoint? When I do the workflow for flux 1D checkpoint it crashes
Same when training with Dreambooth
Can you please share ComfyUI workflow?
Also you bellow the model in Gallery you can filter by "Has Metadata" and tool = ComfyUI to find some example workflows.
@maracon thanks didn't know about that
most of my uploaded images have metadata included, just drag them into comfyui directly from civitai. they are pretty much all txt2img with my templates, but it can be a start. more complicated workflows will come in the future with also advice on generation. for now as recommended, it is full of creators who share their workflows.
@ALIENHAZE Is FLUX model resisting generating realistic female bodies? Lot's of them looks good old plastic way :-) Could you please share your thoughts why it behaves like that?
yes, we are so used to XL's finetunes, that many flux results do not seem up to par. especially for naked bodies. but be careful not to get fooled by the young age of the model. there will be maybe 10 valid photorealistic models in Flux against the hundreds present in XL. also in general there are many more generations correct and indistinguishable from real textures, compared to what happened with XL. Finally, the model itself also started from a much less censored and much less family friendly dataset (in XL they had thrown in everything they had without filters). this, together with the size of the dataset make the finetuning or photographic overtraining process much longer.
In my opinion XL is not dead yet (see 'Pony' still unsurpassed in its field), especially for those who do not have many resources in the computer, but if a new update does not come out, in a few months Flux will completely outclass it, even for those who have few resources with the quantized versions.
@ALIENHAZE thank you for clear informative reply!
Appreciate you great effort in the field. Looking forward for more models from you 👍
Can you share key tags or a guide for your checkpoint please? I want to get more accurate results.
The flux.1 model is awesome. Works great on my 2070 super.
thanks for the feedback. 8gb or 12gb? how much more or less for a generation?
Unet means it will only win comfyUI right? not in forge webui
works on forge also
@DukeNukem47 the unet is basically the pure model without the presence of clips or vae encoders so as to give more freedom of use and have leaner models. for simplicity on most interfaces you can use the AIO version (which already has clips and vae included) as happened up to now for Stable Diffusion.
@jokershb thanks for feedback
Please help - I used this all day yesterday with success. Today, with the exact same settings, this model only generates latent noise which then turns to a grey square in final step. Nothing has changed, what is going on? Automatic1111
if you are using it locally nothing can have happened to the model! either you changed some parameter by mistake or something was updated in automatic1111 that changed the settings. I don't use automatic often so I can't tell you at the moment, I'll try to investigate and update you!
I got this too, not sure whats happened.
I got this too
@notapotatojs998 @exlartechadm936 @yelimeng006368
UPDATE!
I investigated and Automatic1111 still does not support Flux as a model. The closest interface alternative is ForgeUI (https://github.com/lllyasviel/stable-diffusion-webui-forge) or you can use the slightly more complex ComfyUI (https://github.com/comfyanonymous/ComfyUI).
Hi there...this looks great...i have a question. Should i use this model to train a NSFW lora or better use the default one and later use this one ? thanks
I'm wondering the same thing
@tupu @Marcellusk466 I personally use the basic model during training but mostly for experimentation. I can collect more data on the effectiveness of the various elements in play (images, captions, resolutions, parameters and so on.).
I don't have enough elements to be able to give you an answer. Experiment and if necessary give me feedback, so that I can intervene in corrections if necessary.
Why is pruned larger than full?
it's all written in the description! it's just to be able to keep the 'unet' as the main model.
You do not have CLIP state dict! How do I fix?
are you using the AIO version or the Unet? because in the second case you have to load clips and vae separately.
AssertionError: You do not have CLIP state dict!
I thought clip is included. Added the file to "Forge\webui\models\Stable-diffusion". Did i miss something`?
you have to add clip to (taking clues from your filepath) Forge\webui\models\text_encoder
@antonfawkes33350 where take it, pls?
@devastattorr Sorry, what?
@devastattorrIf you are using Forge webui place in tokenizer folder
I still have issues to get hands right most of the time. Any hints?
For now, negative and Loras, I'm working to fix the lack. Specifically, what are the major defects? missing hand, extra or missing fingers, crooked fingers, fusion of limbs...? thanks for the feedback.
@ALIENHAZE Just too many or too less fingers. 1/20 weirdly big or small hands. Many also just with non uniform fingers. So one is too long or too short. I'm generating 100's of images of images at a time and 1/4 have that finger issue.
I'm usually doing fully body shots so the hands and fingers are seen from some distance.
extremely bad quality..
Very nice model!
Any chance for an LCM version? (Lightning/Turbo). I tried the LCM LoRA for SDXL and generates some bad artifacts with this model. My old RX5500 XT board takes quite long for 20 steps, hopping I get it down to 4.
Aren't fp16 unet files ~24GB usually?
Hello, I downloaded several of your models to train FLUIX LoRA, but whether I use the 11GB or 22GB version of STOIQO Afrodite or STOIQO NewReality, I always encounter errors during training. The error is as follows:
Traceback (most recent call last)
File "IAIlora-scripts-v1.8.5/scripts/dev/flux_train_network.py", line 519, in <module>
trainer.train(args)
File "IAIlora-scripts-v1.8.5/scripts/dev/train_network.py", line 402, in train
self.cache_text_encoder_outputs_if_needed(args, accelerator, unet, vae, text_encoders, train_dataset_group, weight_dtype)
File "IAIlora-scripts-v1.8.5/scripts/dev/flux_train_network.py", line 212, in cache_text_encoder_outputs_if_needed
unet.to(cpu)
File "IAIlora-scripts-v1.8.5/python/lib/site-packages/torch/nn/modules/module.py", line 1174, in to
return self._apply(convert)
File "IAIlora-scripts-v1.8.5/python/lib/site-packages/torch/nn/modules/module.py", line 780, in _apply
module._apply(fn)
File "IAIlora-scripts-v1.8.5/python/lib/site-packages/torch/nn/modules/module.py", line 805, in _apply
param_applied = fn(param)
File "IAIlora-scripts-v1.8.5/python/lib/site-packages/torch/nn/modules/module.py", line 1167, in convert
raise NotImplementedError(
NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device.
123000-305770 ERROR Training failed
I’d like to ask if these models are unsupported for FLUX LoRA training?
I see lots mature video. which tool did you use? I'm not sure that kling or Runway will accept it
mmm
Eh, it's okay. It seems like it was trained on SDXL/Pony images, or involves a merge with a model like that, so it looks like what we've already been seeing the past year or so. Very round and sometimes exaggerated breasts, very circular nipples, not crazy detailed skin. Nothing new like Flux is capable of. Getting a topless output from Flux looks like natural breasts but with blurred nipples, but the breasts look like they properly fit the character. This one looks like more of the same SDXL level stuff, which if you like that it's great. Most finetunes seem to have followed this route, but there's a slim few out there like FluxFlux or GetPhat that have a more natural and photorealistic look like you'd get out of Flux. But unfortunately those also tend to get fried when using Loras, but Afrodite is much more Lora friendly. So it is what it is at this point- it took a while for SDXL to get better, so I haven't been expecting miracles yet.
Very helpful and constructive review
Amazing! Need fp8 version
GGUF please :D
Yeah, I'm going to jump on the bandwagon and say this model is absolutely amazing, and I've used it to add some really great details in img2img, but it sure is right there at the limits of what my 4090 can do. An FP8 or Quantized version would be most appreciated.
HI, is there an FP8 version of this model available? I see the FP16 version, but I need FP8 for my project.
Hey! I can't generate images nsfw with penis and hardcore nsfw,just basic same as original FLUX.dev.
Any suggestions where i might be wrong?
Hi hi. Next update?
fp8, PLEASE!!
Very good Model. Thanks alot
Such a great versatile model! How comes it is not listed when you search for Flux checkpoints?!
It's not a true xxx model if it cant do penis, ripped, muscles, and everything mixed. Can it do Futa?
It's not about the makers prefrence, it's about the capability .... lift your game
No, it's Afrodite, as in the of beauty goddess. It does what it was designed to do. Penis, Muscle, Futa have nothing to with Afrodite. It absolutely is about the maker's preference, model training isn't design by committee. What are you even on about
Niche stuff = LORAS
lol, go to Rule34 its spammed with Futa like 80%. Glad i could help you ;).
Leans toward a certain body-type too much.
If in clip be "NSFW" you receive not img
Afrodite XL v1 is the same as this one:
https://civitai.com/models/161068?modelVersionId=219222
Doesn't work in Forge in mode img2img
Hello, thank you very much for the many hours you invested in creating this model. /The best part: I see a field for negative prompts./As I've seen in some tests, it works great for me and gives me natural-looking Flux-quality photos. . I missed this option in InvokeAi v5.10.1 when working with pure Flux. Flux LoRAs don't work with this model, but some others do. Have joy and make good times wherever you go.
Sehr schön! Gelungene Arbeit!
Ich habe einige bewährte Prompts ausprobiert und bis auf ein paar Feinheiten werden absolut tolle Bilder generiert.
Es wäre schön, wenn Hände und Füße nicht immer am Rand abgeschnitten würden. ich weiß nicht, ob sich das über das Model korrigieren lässt!
why is the pruned model bigger than the full one?
Because they were mis-labled? is my guess. I am getting both.
I've been out of SD for a while, so dumb question - where do you get Euler+ Beta sampler? Or did you mean Euler and Beta together?
"Euler" is the sampler, "beta" is the scheduler
Any update to this model expected soon? This is my daily driver... thank you.
Using the SDXL version and it is extremely limited in what it can do compared to STOIQO's own New Reality. Also routinely struggles with hands/feet/fingers, ignores clothing instructions in prompts routinely. TO summarize it can't do porn, has anatomy problems to do portraits effectively, can't do good landscape shots. Just skip this one and use New Reality, WildCard,Cyberrealistic or EpicRealism instead for SDXL.
Are there any examples for Comfyui/workflow?
Is there any workflow already set up to use in comfyUI?
Just Use the default Flux Workflow
@Dead_Sec cany u share the link
@dipi199201 drag and drop this image in comfui GUI
@Dead_Sec Damn, i wish i knew this earlier, it's so easy to just copy someone else's workflow this way... thank you!
@dopa377 mention not ! feel free to ask anything that i can help you with xD
Is there anything special about the t5xxl_fp16, clipul, and VAE of aio version?
It`s better than nsfw loras, but still penises are massacred
never any love for penises or ass holes, I feel bad for any gay men trying to use AI...
I dint understand... so, I need to download the 2 versions? (10 and 20GB each) or just the heaviest one, if so, how do I combien the two of them?
^ anyone, pleaase.
^^
clicl (2 files) on right side
Ich habe verschiedene Modelle verglichen. Dieses hier ist für mich immer noch das beste.
Details
Files
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
Mirrors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
FLUX_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXSD_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
flux.1.dev (STOIQOAfroditeFLUXXL_F1DAlpha).safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
Mirrors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha_1.safetensors
STOIQOAfroditeFLUXXL0F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha_Prunned.safetensors
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.