IG baddie all about the good life
Description
Comments (4)
This is actually a great Character LORA. Thanks 4 sharing. The previews are quite blurry, though - almost missed this one.
Well, i just added a nice fashion show gallery, hope you like it!
Maybe important to add, if nothing like "black girl" (or maybe other skin color triggers) was prompted, i actually got an unrelated white woman in one scene.
For the actual LORA test results, i had some slight inconsistencies like the sometimes patchy skin, but it could be the simplistic prompt settings i used. Jade model reacts amazingly to poses, expressions, clothing/accessories, at least the lingerie i tested.
As a bonus, i also tried some changes by adding e.g. "with a soft-voluptuous body" (balanced), "a voluptuous african-american body" (more booty), "a stunning BBW body with massive natural breasts". It looks great but isn't consistent between prompts.
Most importantly it rendered very sharp and realistic. Well done.
Thank you for the input i am really new to this and learn all i can. Took me forever to learn how to make this LoRA but i any tips on making pics would be great and very open to them all. thank you
Hmm okay so i just started to write a lenghty, detailed guide on how i use different methods to achieve great renders. But maybe that's not entirely necessary.
So when i look at your previews, they don't show any info about the used settings at all. They do look very blurry with a strong bloom effect and glossy skin. While my images from the fashion shot look perfectly crisp and natural, like ultra-HD photographs.
Means, you maybe just need to render with different settings, and maybe a different checkpoint. If you click on any of the images i uploaded here for your LORA, the right tab shows all relevant info - checkpoint used, resolution etc - except the tool (Forge NEO) and the Scheduler (Beta as per Sick Ollie checkpoint).
So just try that. If you want, can also write a longer post, about the mentioned different methods - minimal prompts with great effect, longer prompts and what they need to work. I also use LLM AIs to write prompts for me, either for full prompts or just how to describe details in a way Krea understands.
But for now i think it's your render settings and methods.
Okay i see the settings on your picture now.
Krea2 needs cfg effectively disabled (or it's at 1 in Forge Neo, cannot go lower, not sure how ComfyUI handles this), this also turns off negative prompts (which Krea doesn't use).
Steps are usually 8 with Euler or ERE SDE as sampler, and then Simple or Beta as Scheduler. Sick Ollie suggests Euler-Beta at 9 steps, but you can for example use 10 steps if you get a six-finger-hand on a rare occasion (the extra finger should be erased with extra steps).
1024x1024 would be considered the absolute minimum resolution (for a ~1 megapixel image), but most people don't go below 1024x1536 (2:3) or 1280x1600 (Instagram 4:5 standard) for quality.
Best images are probably with 1280x1920 (2:3) or 1536x1920 (Instagram 4:5) if your machine can handle that.
Standard frame size here on civitai also seems to be 4:5 but most people use taller dimensions to hog the screen space for attention. If you want the Instagram tall resolution, it's 9:16 @1080x1920. Instagram wide is 1.91:1 @2048x1072. Using different dimensions than 4:5 will auto-crop images to that size in some windows, though.