This is fucktastic. A porn-generating machine for Pony/PDXL.
It's focus is on creating high quality, photorealistic, but exaggerated images with a focus on women.
Also available on mage.space.
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.
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.
Recommended Settings
Steps: 20
CFG-scale: 7
Sampler: Euler A
Scheduler: Beta
Description
Holy shit this version hits different!
FAQ
Comments (16)
Hands down one of my favorite models. Unbelievable results with it. Can't wait for 5.1!
Not sure, if I like the new direction turning models into LCM schedulers - most LCM based look very similar in results.
These models do not have any LCM training, this is ran on top with DMD2 LoRA.
If this is what you mean.
Not true what OP is saying I used this model in ComfyUI and I am getting the same results like the examples and generated images by users, all depends on how you configure your ComfyUI and what nodes you have in your workflow
Hello! Am I right, that using more steps gives higher quality? How much steps will be most effective for this model? And does LoRa/DoRa make some bad difference for steps more than 50?
"It Depends", for all of the above XD
usually no, just use the XYZ script: use "steps" on one axis and then try, 20,30,40,50 . I usually dont see improvements above 20-30 these steps rules come from 2023 SD1.5 Times imho. XL Models seem to work only if you use a 2 pass workflow (ie Highresfix). Thats where the quality comes from not from the steps.Also Depends on samplers, you can use the XYZ script to find samplers for the models you use by trying alls samplers with one good sample image and same seed (dont use loras for this test since some can skew the test in my experience)
It depends on the sampler. Generally Euler A caps out at about 30 - 40 steps (according to A1111 docs), but samplers like DPM usually get better the more steps you use, with some people suggesting you should be using around 60 or more, but I think the generation time compaired to the quality increase doesn't make it worth your while. As @pink0909 suggested, if you're still unsure just try an xyz grid comparision
5.1 is great. Thank you for sharing such a beautiful model, and making it free.
Any chan e this coming back online?
Sure, anyone can pay the crazy amounts asked right now to put the checkpoint on the generator for a week. Just press that bid button and say goodbye to maybe $20?
Hey thanks for your work, can you or someone else tell me how to create consistent character with this model? I would like to create an album for train new lora to make my character, but every time there is differents details.
Thanks for the support :)