Welcome to Perfection Amateur!
This is an Illustrious XL merge with a bunch of hand-selected checkpoints and LoRAs to improve the base model's realistic capabilites.
This is a special branch-off for cinematic flavor and for when you need your images to be extra juicy and well-lit!
It's focus is on creating high quality, photorealistic but amateur style images with a focus on women.
Each sample/preview image contains the used workflow. Here's a quick article with simpler more beginner-friendly workflow. This is a recommended starting point.
Do not use the Normal scheduler. Use Simple or Beta!
Use euler_a, dpmpp_2m_sde or dpmpp_3m_sde as your sampler.
To use the faster low-step settings, you need to use the DMD2 LoRA. This allows you to use 8 or 16 steps and generate very fast. Remember to generate in 2 passes, either using "Hires fix" in Forge/A1111, or by creating and chaining a second sampler node in ComfyUI. A sample workflow if included in the sample images of the model. Use this as a reference.
Recommended Settings
Base Generation
Steps: 30
CFG-scale: 3
Sampler: DPM++ 3M SDE
Scheduler: Simple (beta upscale scheduler)
Upscale/Refine
Steps: 16
CFG-scale: 4
Sampler: DPM++ 3M SDE
Scheduler: Simple
Denoise: 0.35
Description
Credits to Pony Amateur ✨ by ZyloO
FAQ
Comments (35)
The worst thing is, it should be piss easy to prompt sex with illustrious, but the XL model destroys every adherence.
Image generation is not working, the images are broken. I'm using Macbook Pro M4, is because this model uses float16? Please help.
I have been trying to use this checkpoint and I keep getting odd paisley and flower patterns in my backgrounds. Seems to show up a lot on walls, couches, and sheets. Any way to combat this?
Edit: I had a discussion with Chat GPT about it and apparently some certain phrases have a adverse impact on image gen. For instance "warm tones", using this phrase you are probably meaning warm light, maybe firelight, candle light, etc. Some models interpret warm tones to mean "a picture in a 1972 whorehouse attic, include the tacky paisleys and antique furniture". From what GPT says this is from some models being trained on.....1970's erotic photos and having that term associated with those photos. I removed that term and it mostly got rid of the patterns.
I think there's an issue with the uploaded version, clearly isn't the one used in the civitai generator.
@conorqangel48532 It's 100% the same.
You can download the image and drag it into ComfyUI and generate, and you'll get the same image.
@Darvidous Interesting, I haven't quite seen this, could you share some generations to check it out?
Maybe it's overfit on this pattern.
I get the same pattern on everything. Wallpaper, bedding, etc.
yup i get the same thing. i generated like 3 separate images today that produced this pattern. anyone have any negatives that they found filter it out pretty well?
Thanks for sharing.
May i know the main differences between 2.0 and 2.2?
It includes additional Amateur camera / style type LoRAs merged in to go even stronger in this aesthetic. This is one of them: - https://civitai.com/models/480835
Thanks for sharing, it seems great. I just keep getting blurry pictures, anyone with a fix?
Can you tell me about your settings?
Which program are you using to generate?
@6tZ i'm having the same issue even using your prompts, this is my settings: https://prnt.sc/h7V8N77_1ac5
@Leamas_ Good, so you get the first pass well. Now you need to upscale it. If you look at the workflow in the images of the model, you'll see that we take this output and scale up the latents to 2x, and then run another sampler at a ~0.15 denoise level, and similar other settings. This is what adds the quality to the image.
This first pass creates the shapes and general composition, and then the second step adds the quality and details.
Additionally, you need to load the DMD2 LoRA for this.
https://civitai.com/models/1608870/dmd2-speed-lora-sdxl-pony-illustrious
This is what makes the low step count actually look good. If you don't use that LoRA it won't work well, and it'll just be a blurry mess.
@6tZ got it, thx bro
is there a LORA that will make it look more like SDXL realistic instead of the uncanny Illustrious/pony realistic?
If you're using ComfyUI you can always merge SDXL back into the model.
@6tZ i'm not, but I've made a few merges in forge (its easy in forge, i have no idea how you would do that in comfy)
@SLACK69 In Comfy you just manipulate the model, such as load the LoRA, then save the model with a model saving node, or a checkpoint saving node, depending on the format you want to use it for later.
For Flux, you use the "Save Model" node and save only the model. For SDXL etc. you use the "Save Checkpoint" node and include VAE/Clip.
There's different merge nodes included, like "Simple Model Merge" or "Simple Clip Merge", which are ratio based and easy to use. I recommend starting there.
@6tZ so you can merge LoRA's with models and other LoRA's?
@SLACK69 Yes you can. You may need to do more merging, like subtracting original models and such.
v2.2 Only seems to generate either extremely low quality images or random noise on Comfy. Even loading the workflow directly from other images just generates random noise. Can anyone else confirm? I suspect maybe the uploaded version for local download is borked somehow.
Let me download and test the model on the site. Maybe it got corrupted somehow...
I do see only a few user generated images below.
Nope, it works fine with the downloaded model.
Are you using the same settings as shown on the model description and in the provided images/workflows?
Try the workflows attached to the images.
You can simplify and remove the multiple prompts and ignore the save nodes and whatever.
Focus around the 2-tier approach to generating a low resolution first, then upscaling, with the provided settings. Also written in the model description.
https://civitai.com/posts/19476567
The final look is SO good!
But wow it messes up hard when it comes to NSFW multi character scenes.
Can someone teach how use civitai models and where to use I don't know help me brother's 😭
If you want to use them locally on your own computer, search youtube and learn how to install and use tools like ComfyUI. This is the best developed platform to run and use image generation models.
If you want to use them on the website itself, it depends a bit.
There are "generally" 2 types of models you'll find on CivitAI.
"Checkpoints / Base Models"
LoRAs
For checkpoints, you are only able to use the 200 most popular models each month. Popularity is voted for/paid using an on-site currency each week. So for example, this model is a checkpoint, and is not able to be used until someone votes for it by paying to be in the top 200. It costs around $10 to do that for a week.
For LoRAs, these can be thought of as a "plugin model", which you can use on top of other models, provided you choose the correct type (architecture), of a base model. These can all be used on CivitAI.
You generate by pressing the "Generate"-button on a model page. If you don't see such a button, it's not available for on-site generation.
You can read the website documentation to learn more.
6tZ Forge or Reforge is much better suited for beginners and those who don't plan to delve too deeply into all this.
anonimus12345678902 Yeah, but it doesn't support Forge, right?
6tZ I use both ComfyUI and Forge with the help of Stability Matrix. Both work fine.
Get Pinokio and use that to install an AI art program like Comfy.
Otherwise you're going to be copying and pasting random bits of code that you don't understand into the command prompt.
Youtube it you goob
how to load lora on this workflow? need to get that big bewbs yo
Just add a LoRA loader node, or use the built-in prompt lora loader.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.