If you enjoy my contribution to this community, feel free to buy me a coffee: the more caffeine I drink the more models I can create 😅
Comparison gallery here: Fv6-Fv5 and NSFW illustration comparison between Gv1-Cv6-Fv2.

Current SOTA model in my experiment:
SDXL Base model: Gv4 is the most balanced model allowing both realistic and styled NSFW and SFW images. Better aesthetic than Fv6 but less photorealistic.
SDXL photorealist (SFW and NSFW) model: Fv6 is the way to go for hyper-realism including realistic NSFW images but it mostly lacks the styling capabilities of Gv4.
SDXL Illustration : Gv4 (SFW and NSFW). Cv6 is however still worth a try if you are not into NSFW images.
Flux model: Flux1-A1
Image Generation Settings for SDXL models
DPM++ 2/3M SDE / Karras or Exponential are always a good bet with 25+ steps and CFG around 5-7. But DPM++ SDE / Karras with less steps (e.g. 12) and higher CFG (8-11) is worth a try as is Euler Ancestral / Normal for a less detailed result which may also be interesting with a model like Fv6 to get something less "noisy".
The default CLIP Skip of 2 is also a good bet, but using 1 or 3-4 is also worth trying: 1 push more towards prompt adherence and 3-4 give sometimes a better result than the default focusing more on the "concepts".
Image Generation Settings for Flux models
My preferred settings are DPM++ 2M / beta or sgm_uniform or DDEIS / normal for the sampler / scheduler, beta giving a bolder stronger image. For a more subtle image, Euler / simple or beta seems a good bet.
CFG seems to have a huge impact on the final image and be very sensitive even to small variations.
For photos, CFG should remain low (1.5-2.5) to avoid plastic skin.
For fine art and illustration it is more complicated because it depends on the medium. For "rough" styles (painting, watercolours etc.), CFG should stay quite low in the 1.5-2.5 range but for anime or comic style, CFG needs often to be pushed further to achieve the desired style (3-6 or more).
If the image is messy/malformed or blurred, it is often because the CFG/steps are inappropriate for this image, but it is not always easy to know whether CFG/steps must be increased or decreased (at least to me 😊).
There is for sure a lot to learn concerning Flux behaviour which is quite different than SDXL and we will need to adapt.
Workflow
All my images are uploaded with embedded ComfyUI workflow which is alas incompatible with CivitAI processing and most often prompt cannot be retrieved. You can however download the original PNG image with the workflow included by clicking on the "DOWNLOAD" icon in the image viewer.
Why I try to mostly publish images straight from my model with maybe a bit of a Lora (mine or some detail enhancer), I also sometimes now use Controlnet to get better more detailed compositions more easily: in this case the source image is obviously not in the workflow, but I guess you can still use the image I publish as a source if you want to make a variation :)
Past
Starting from E serie, models are evolving sometimes by merging with other models (thanks to other contributors!), but mostly via training on my own dataset: a modest dataset (~2000 images currently), but I try to somehow compensate with quality and originality.
Starting with Fv1, I have included many synthetic images I created using previous versions: playing hard with prompt and retouching when necessary the result in Photoshop in order to have a dataset which contains many original images.
The core idea behind this model was to create a versatile tool by merging some of the best existing models which fits my personal taste (photography and fantasy art to make it simple). My primary goals were:
Photorealism: The ability to produce stunningly realistic images of both people and objects/nature.
Flexibility: The ability to create highly stylized images, allowing for artistic expression through various styles and combinations of artists. I am from a older generation and comes from Europe so "style" does not mean for me "Japanese kawaii aime with boobs" or "DC Comics cartoon with lots of superheroes and voluptuous blonde babes" but more from a univers of Frank Frazetta, Milo Manara, Boris Vallejo, H.R.Giger, Wojtek Siudmak and such fantasy art masters: there are boobs involved for sure, but the style is somewhat different :P
As I doesn't like being limited in my exploration of the human body, the idea is also to have a fairly capable NSFW model. However, dur to the nature of available training images in the data sets, NSFW often comes with a strong bias toward either porn photos or porn Japanese anime and impact the flexibility (typically as soon as you use the word "sexy" in your prompt you need to weight in the style). This point is thus NOT the priority for the base model but is pushed forward in the NSFW model.
Description
Here it is : my very first Flux fine tuning!
Flux.1 is really quite an amazing foundation model and it seems that Lora are very efficient, so I did use the dataset I built for my SDXL model to create several Loras specialised on specific concepts and then merged them with Flux-Dev-FP8. A few available Loras available here have also been included such as FaeTasticDetails.
This version is so a "Dev-FP8" model which includes CLIP and VAE.
It obviously has to be considered as a very early model and the journey is just starting :)
FAQ
Comments (39)
Excited to see you've made a Flux checkpoint!
Question, I usually run the dual clip loader in Comfy separately instead of using Clip from Checkpoint Loader. Is that OK with this model or did you train something specific with Clip?
Flux is too good to ignore :)
To be frank I am still not completely sure how it works: I trained Loras with captions using AI Toolkit by Ostris and I don't think that the text encoder is trained at all so the Clip of the merge should be identical to the base model. So I think it should work just fine with the Clip loader... let me know if you encounter some issues!
@aramintastudio I agree with you on Flux!
So far in my tests I've tried with/without separate Clips. But in both cases faces don't come out as "crisp" compared to base Dev, and the images have a higher tendency to break down sooner with the "screen door/grid" artifacting that we see in base Flux Dev as it approaches 2000px/side.
I am only giving you this feedback to help identify weak points; I know it's a learning process and experiment, and I appreciate your work! Your SDXL models were consistently some of my faves so I am sure you'll also get there with Flux :)
@EnragedAntelope Thanks for the info: I tested with "standard" 832x1216 size mostly and I agree that there is a lot to learn about Flux behaviour and how to fine tune it. Personally I am a bit concerned with the non-realist (illustration) part as it seems Flux is a bit a hit or miss on this point. Anyway, thanks for your feedback!
Please upload just the Unet model without the Clips and Vae if possible.
I am still in the learning phase of how it works and I don't know hot to do it :/ Will have a look.
@aramintastudio hi, here's a lead https://www.reddit.com/r/StableDiffusion/comments/1eukgax/unet_extractor_and_remover_for_stable_diffusion/
I'm just going to second this, and thank volatilevisage2 for linking one way of decoupling the unet from the clip and vae. This is going to be important as we now have 5+gb of clip being duplicated instead of ~160mb on SDXL.
GGUF version plz
I have barely heard about GGUF ... things are going too fast :P Will have a look!
Glad to hear that. Your example and model for other version looks fantastic! If you could make GGUF version, more user with 12GB VRAM or less (like me) will have a good time with your creations.
The Flux models are so large !
Can you also share the loras used to create your model?
Sorry, but these are not Loras which would make any sense as standalone Loras as I trained them only with the purpose of merging with Flux to achieve certain goals.
Hi! Any chance of a Flux NF4 version?
People ask for many different versions and I don't even know at this point how to convert my model to those versions :) But A1 will not last long I think and for now I will concentrate on making a better A2 version probably.
@aramintastudio You can create a nf4 version in Forge. See here: https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1224#discussioncomment-10384104
Araminta for SDXL was already one of my favorites for that model, but now one for Flux.D? Be still my beating heart - very excited to see!
Thanks!
But end result is still unknown with Flux... still in the discovery phase :)
Great results on ForgeUI with Flux version ! thanks a lot !
Don't use Forge myself: glad to hear it works well :)
can you please create a version without the T5 included. There is no need to include text encoder
Congratulations on getting this model out so quickly. Looking good (a couple of images not as sharp as I would have expected from Flux and one gave a very non-descript output which didn't really pick up anything of interest in the prompt but generally excellent results). I then ran a test to compare the default flux1-dev-bnb-nf4-v2 model with Araminta F1-1A (plus a few images in EV4 & FV1) using Forge; all based upon British cars - this also proved a good test of AI as to whether I could get cars driving on the left with driver and steering wheel on the right and correct number plates! You can see the results at: https://civitai.com/posts/5659086, https://civitai.com/posts/5659431, https://civitai.com/posts/5659146, https://civitai.com/posts/5659210.
Some of the comparison images use the same seed.
Thanks for the feedback and the tests!
Maybe it was too quick :) There are actually obvious flaws in this first version (especially concerning non-realistic styles which are mostly lost) and the sometimes blurred images indeed.
I am already working on a different approach and it should be even better soon ^-^
FP8 supposedly suffer some quality issues over FP16, can anyone who runs both models Flux1D a lot confirm?
The difference is not enormous. But yeah there is a dip in Quality when using "fp8" mostly in Prompt understanding and composition.
can you please tell us what sorta loras you trained flux on cause the main thing i've been looking forward to are artist and photographer keywords finally being trained into flux finetunes. Also please can we get a unet only version without text encoders and stuff.
I didn't train on artists/photographers names, only generic concepts and overall styles such as glamour photography and exclusively photos. I miss also using the artists/photographers names as I do in SDXL, but I am not sure it will be as easy with Flux to have them back as SDLX and even SD3 know those artists styles at their core which doesn't seems to be the case with Flux.
For the unet, I will wait for another version because A1 is really a WIP and I am already trying to get something better. But at this point, it doesn't seems to work as well as it did with SDXL :/
@aramintastudio What is it struggling with?
@MrHYD With SDXL I were evolving the model incrementally by merging with other models or Loras and the model were becoming better and better without any degradation. With Flux, while it seems to work fine for realistic images things go wrong quickly with illustrations or text for example and the model quickly loose the ability to create non-realistic images. But it may be an issue with the way I merge : will need to investigate that.
Seriously one of the best models I've used. Download it. You will never regret it.
Thank you! 🤩
flux first harcore model, we must add at the end of the prompt to get a good quality "(ultra detailed: 1.1), 8k, UHD:1.1"
Use fooocus options,"Enhance"
Очень хорошая модель,я не понимаю почему так мало реакций на неё, она превосходит все что я видел
works great in Ruined Fooocus 1.55 ! thanks a lot !
great vaginas my dude 10/10
edit: this model is amazingly coherent AND compliant; I can't believe how many concepts it can properly depict.
What do we need to get A1 working on Forge? Thanks if anyone can help.
So many good things to say about this model, but the skin texture in particular, is exceptional,
One of the best models on this site, and I mean for real! Amazing nudity details.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.















