SmthMix_Semi-Real -
is a semi-realistic model is the result of my ideas about what kind of model I wanted to create. Which does not delve far into realism and is not too anime.
Recommendations For Use:
Resolution: 832 x 1216 - 960 x 1520 (You can use others as well)
Samplre: Euler A or DPM ++ 2M SDE and Karras (You can also try DPM2 and Exponential)
Steps: 25+
CFG: 5
Clip skip: 2
I also use highres.fix Upscale
Positive prompt
-masterpiece, best qualityNegative prompt
-low resolution, low quality, bad quality, worst quality, signature, watermark, lowresTensorART - https://tensor.art/models/904330146402562129/SmthMix_illustious-v4
SeaART - SmthMix_Semi-Real - SeaArt AI Model
Description
FAQ
Comments (5)
Great checkpoint，But there is a question, does this checkpoint have higher requirements for video memory? Once Hires.fix is used, the video memory is exhausted, and it does not work even at a magnification of 1.1. The same image resolution can be generated normally when using waiNSFW with Hires.fix magnification of 1.5. My basic resolution is 1024x1536..
Hello. Hmm, that's strange. It shouldn't be like this. I also create images using Hires.fix and have no problems. I don’t even know what the issue could be. Can you send me the image you want to create with upscaling? I will try to create it on my end.
I just launched a test. As a result, the same prompt and settings with Hires.fix in my model and WAI v14 created images at the same speed. About 3 minutes.
Hi, great model, I noticed the model seems to have trouble recognizing the costumes of different characters. When given the prompt it illustrates the facial features accurately, but not their costumes. Is it trained that way?
Hello. It's quite possible, as when creating the model, I didn't pay attention to how it creates different characters. I was just aiming for a certain style.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.