I wanted to try a realistic merge using the same philosophy as One-Shot Illustrious. I've found realistic models to be very difficult to work with and I'm not sure I was successful, but I decided to share anyways in case someone might enjoy using it. Check the suggested resources below for models used in the merge.
All example images were generated without loras and without upscaling to show the natural output of the model. This model has an Asian bias, but it should be possible to generate other ethnicities by adding related tags to the positive and negative prompt.
DISCLAIMER
While this model can generate SFW images, it is NSFW and is capable of generating explicit material. Use caution in your prompts if you're looking to avoid anything NSFW, but I can't guarantee a random NSFW image won't appear from time to time.
RECOMMENDED SETTINGS
These are the settings I use, but there could be something better. Feel free to experiment on your own.
Resolutions: 832x1216 / 768x1344 / 1024x1536 <- what I use (Caution: Higher resolutions may increase the chance of incorrect finger and toe count, as well as other oddities. If you still want a higher native resolution, 1248x1824 has been surprisingly okay during my limited testing.)
Steps: ~35
CFG Scale: 7
Sampler: Euler Ancestral / DDIM
Scheduler: Karras
Upscaler / Hires. fix (Optional): 0.3 strength, 2x upscale, 4x-Ultrasharp or SwinIR_4x
ADetailer / CLIPSeg (Optional): Use as desired/needed
RECOMMENDED BASE PROMPTS
Same as above, this is just what I use. Feel free to experiment to your own preference.
Positive: masterpiece, best quality, amazing quality, detailed eyes, realistic, photorealistic
Negative: bad quality, worst quality, worst detail, sketch, censor, bar censor, extra fingers, extra toes
If you don't mind sacrificing realism for better prompt adherence (or if you just prefer the look), remove "realistic, photorealistic" from the positive prompt. If you're trying to generate an image without a visible face, be sure to change "detailed eyes" to "detailed" in the positive prompt.
Description
FAQ
Comments (5)
really great checkpoint.
Does what your description says, hands, fingers, eyes, all seem to look good!
(using fooocus with cfg 7, dpmpp 3m sde gpu, sgm_uniform)
How are these checkpoints when it comes to generating images without people in them, like objects, empty rooms or landscapes?
For images without people, you will need to combine the "no humans" tag with at least one other non-specific environment descriptor tag such as "scenery" or "interior". The checkpoint is pretty biased toward showing a person, so using only one or the other is not enough. For example, if you wanted to show a bedroom without people, you will want to add "no humans, interior, bedroom" at least.
I don't normally generate these types of images, so I'm not sure how it compares with other checkpoints. Depending on what you're trying to get, you might need to use checkpoints that are more tailored to environments.
Why does it called one step if it's not capable of generating at accelerated rate? I got clickbaited
I'm sorry if it seems like clickbait. The idea for the name is conceptual and not meant to be related to any technical aspect of image generation, so it's unintentional.
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.