- This model was converted to Core ML for use on Apple Silicon devices. Conversion instructions can be found here.
- Provide the model to an app such as Mochi Diffusion Github - Discord to generate images.
split_einsumversion is compatible with all compute unit options including Neural Engine.
originalversion is only compatible with CPU & GPU option.
vaetagged files have a vae embedded into the model. Most common are 1.5 or 2.1 others are named accordingly.
- Descriptions are posted as-is from original model source. Not all features and/or results may be available in CoreML format.
Note: Some models do not have the unet split into chunks.
PLEASE READ DESCRIPTION
Deprecated model, consider updating to: https://civitai.com/models/5935/liberty it is also licenseless
Very NSFW biased model!!
-aErosNoVAE: It is the exact same file as aEros prunned but without the VAE requirement, which also means that you can now train embeddings succesfully with aEros. (VAE would still be recommended for normal use).
UPDATE: This model has experimented a restrictive change of license ONLY because it inherited it from some of it's building models.
Until further notice, I am not developing it anymore.
it needs vae-ft-mse-840000-ema-pruned (or another one if you want to get experimental) or it will output broken images.
Art&Eros is out!!! This is my second model. It is similar to REA (my previous model) in it's concept, and many of it's building materials are the same, so most that worked for REA works for aEros, but it is not an evolution of it. It has been rebuild from the ground to be 'better, faster, stronger'.
My thanks to everybody who made it possible, because I if there's anything good on it is because the source material was great too. In no particular order to Hassan, AloeVera, the CivitAI Team, Izuek, Someone88, wavymulder, UnstableDiffusion Team, justmaier, moist, sviasem, kavellion and any other creator I might not have been able to cite.
The main download is the PRUNED safetensors NoVAE version. You have the unpruned ckpt as an optional download.
First it may not be the best begginer checkpoint out there. I consider myself experienced at prompting so I haven't tried much basic prompts, however I doubt that a plain "pretty naked woman, big boobs" is going to take you very far with it. However, although I prompt a bit different than that I have tested that it understands the words from the language of PhotoReal v0.5 (it is contained within the merge), so if you are having troubles getting good outputs from it you can begin from there, but in general, try to use a more natural language than an array of commas: https://docs.google.com/document/d/1-DDIHVbsYfynTp_rsKLu4b2tSQgxtO5F6pNsNla12k0/edit
There's also one screenshot with a very basic example prompt for you to get the idea.
According to my testing it is a very powerful model for it's purpose, but it is a project I made for myself which means that depending on what you want it may or may not lack certain areas. It has been intensively tested to generate photorealistic(ish) images of different types of girls in different poses in different places wearing different things with different artistic moods. Nothing more, nothing less. From hardcore, to to group, to drawing, are out of the scope uses that may or may not work (But have improved since REA). It's community report that it is very good at fantasy and landscapes/interiors too.
List of keywords known: 'elden ring style', 'postapocalypse', 'knollingcase', 'cyborgdiffusion', 'analog style', 'swpunk', 'synthwave', 'dreamlikeart', 'modelshoot style'
'bf' is no longer a trigger word, nor part of the model.
HOW TO USE IT:
You MUST USE vae-ft-mse-840000-ema-pruned (or experimentally other VAEs). Otherwise it breaks.
Some users report having problems using something different than Automatic1111webui. Cannot troubleshoot that myself.
This model DOES NOT REQUIRE TO USE TRIGGER WORDS.
Trigger words are general style or composition modifiers. They can be used alone or in combination and will give an special mood (or mix) to the image.
Trigger words have only been tested using them at the beggining of the prompt.
There is no longer a proper order to mix trigger words between them, needs experimenting for your desired outputs.
Works for a wide variety of steps from 20 to 130 tested. I use 66 Euler A + hires fix.
It works on it's own without triggers at making general or NSFW images of ladies of high quality.
Resolutions tested are 512x512, 384x704, 512x768 and 768x768.
I haven't kept track of all the steps done in the merge (I used a weird methodology), but this is what's inside in different proportions:
Elden Ring Style: https://civitai.com/models/5/elden-ring-style
Analog Diffusion: https://civitai.com/models/1265/analog-diffusion
Hassan's 1.4 and CandyBerry: https://civitai.com/models/1173/hassanblend-all-versions
Modelshoot style: https://civitai.com/models/2147/modelshoot-style
SynthPunk Search: https://civitai.com/models/2856/synthpunk-search
Dreamlike Diffusion: https://civitai.com/models/1274/dreamlike-diffusion-10
Cyborg Diffusion: https://civitai.com/models/1365/cyborg-diffusion
It also works GREAT as an extra style modifier with this hypernetwork for extra artistic outputs: https://civitai.com/models/1141/mjv4-hypernetwork
Honorable mention tu Uber, whose model is not part of the merge, but helped me imrprove general hardcore capacity of aEros: https://civitai.com/models/2661/uber-realistic-porn-merge-urpm
Images are a chaotic trip of LSD colors: You are not using the VAE.
I promiss that I have the VAE!!!: If you swear over the grave of Chanquete, then another user of Discord found the solution, but please, BE SURE THAT YOU HAVE THE VAE WORKING FOR OTHER MODELS:
xtrebel - "In case anyone else is still having trouble with LSD colors, despite having the correct VAE loaded like me... I discovered this checkbox in settings has to be off/unchecked for it to properly load the VAE: [ ] ignore selected VAE for stable diffusion checkpoints that have their own .vae.pt next to them"
Images seem like real images but are a mess of body horror and whatnot: You need to keep working in your prompt. This is NOT an easy to use model (not rocket science either).
This project is considered finished. From now on it is going to become my base model. I may start to train it as the big kids do. But I am more likely gonna try to update it with the updates some of the building materials have experimented. Be it aEros v2 or a totally new thing... Who knows? The sky is the limit!!! Stay tuned on:
- Downloads last month