Any chance for sagging, natural breasts, or only perfect titties be able?

z-image-turbo系列都是对真实人像RL调优，对二次元都不合适。二次元还是看看Noob/illustrious或者去找专门的z-image-anime好点。

If anyone would like to try to reproduce my issue, this is the prompt I'm using (4.3 version):

A 41-year-old mulatto woman stands in the center of a dense autumn forest. She has long curly black hair with loose voluminous curls cascading past her shoulders to mid-back, fair-to-medium warm-toned skin with a subtle golden undertone, delicate yet mature facial features, large dark brown eyes, and full soft pink lips.

She has a softly curvy figure with gentle hourglass proportions: moderately full bust, defined but not extremely narrow waist, rounded hips, fuller thighs, and softly rounded buttocks. Her body shape is noticeable and feminine.

She faces away from the camera with her back turned toward the viewer, head turned to look back over her right shoulder directly at the camera with a calm neutral expression, eyes meeting the viewer's gaze.

She wears very tight medium-blue high-waisted jeans that closely follow the shape of her legs, hips, and buttocks, outlining the gentle curve of her thighs and hips with mild stretch in the denim fabric; subtle tension lines form across the upper thighs and around the hips from the snug fit. The jeans are fully unzipped at the front, metal zipper pulled all the way down; she uses both hands to pull the waistband and open fly of the jeans downward to mid-thigh level, sliding the denim below the fullest part of her buttocks so that her entire buttocks are fully exposed; the top edge of the jeans now sits bunched tightly around the middle of her thighs, creating multiple small horizontal folds and creases where her fingers grip and pull the material downward; the jeans remain stretched taut over her lower thighs below the bunched waistband. Her buttocks are smooth, rounded, and symmetrically shaped with soft natural curves, subtle dimples at the lower back, and a gentle cleft between them; the skin tone is even with a warm medium undertone from the sunlight.

On her feet are glossy dark brown leather thigh-high boots with low block heels; the leather shows slight creases at the knees and ankles from natural movement, reaching two-thirds up her thighs where the boot tops press gently into the exposed skin just above the bunched jeans. She wears a loose-fitting cream-beige cashmere-blend sweater, slightly oversized, with wide neckline exposing the back of her neck and upper shoulders, long sleeves pushed up to mid-forearm, soft fabric draping and folding naturally over her torso and upper back while softly suggesting the moderate fullness of her bust from the side profile.

Forest background: tall maple and oak trees with dense canopy of vibrant red, orange, and golden-yellow leaves; many leaves already fallen, creating thick carpet of warm tones on the ground. Soft late-afternoon sunlight filters through branches in diagonal golden beams, casting long gentle shadows and warm highlights across her back, hair, exposed buttocks and thighs, and clothing. Thin morning mist lingers low between tree trunks. Ground covered in dry leaves, small patches of green moss, scattered acorns. Air appears crisp and cool.

She stands on a narrow dirt path covered in fallen leaves, full rear view from slightly low angle to emphasize the curve of her hips and buttocks, weight shifted slightly to her left leg to create a gentle S-curve in her posture, both hands actively holding and pulling the jeans down to mid-thigh, elbows bent outward, head turned over right shoulder. Natural depth of field with foreground leaves slightly out of focus, sharp focus on her figure, face in profile, hands, and exposed skin.

No text appears anywhere in the image.

Same here. Try describing only the visible parts of the finished scene, and as they already happened. In this case, "jeans are fully unzipped at the front, metal zipper pulled all the way down" would not be visible in this final image. The clothes at their final position, not how they get there, so "sliding the denim" would capture that precise frame, and if I understood correctly, the picture should show the denim already pulled down, so try describing it in that way instead. For good measure, try terms like "mooning". I don't know if ZIT would recognize that tho.

@null I get what you're saying—and I'll give that a try, too—but this exact prompt works almost every time in multiple other checkpoints. This checkpoint gives such great results with everything else I throw at it, it just seems super odd that this particular image is such a struggle.

@97Buckeye the more you tell the model to contradict itself (most often by telling it to show something that it can't show without rejecting other parts of the prompt) the more likely it is to fall apart. different models will fall apart in different ways. so for example it's not just the zipper that's the problem, it's that you've described the jeans in their fully on condition ("follow the shape of her legs, hips, and buttocks") but also ask the model to have the jeans halfway down her legs. the model literally cannot do both at once, so it picks one (or neither, or it goes weird like adding extra bums - i've seen it all). but each model will fail in its own way. sometimes in you favor, by chance, but in this case not.

Use a lora. search "jeans down"

the model goes into diffusion model folder , not checkpoint folder.

@6tZ it does! it's showing now, i had to close my comfyui and load it again, it appeared after a few try

@heatchv01812 it's showing now, i had to close my comfyui and load it again, it appeared after a few try 

Considering the fact that after three months nipples and areolas in all ZIT models still look almost like they were generated by SD1.5 while big breasts still got that lowres grainy jpeg skin texture - that's gonna be a long journey before we finally reach to the proper gentlemen sausages. Even though back in time I completely switched to SDXL only four months after it's release - truly fantastic NSFW checkpoints emerged only year after it's release, NatViS as example.

curious to know wich are your other top 3 models?

@6tZ Really enjoying the model so far! The quality and detail are excellent.

One thing I’d love to see in future versions is a bit more variety in body types. Right now it leans very heavily toward tall, idealized proportions and perky breasts. It would be amazing if the model could also handle shorter stature, natural saggy breasts, softer bellies, cellulite, and other realistic imperfections while keeping that same high level of realism and sexiness.

For me, the ability to generate truly natural, imperfect female bodies is one of the biggest strengths a good NSFW model can have. Would love to see Z-Image improve in that direction.

Keep up the great work! 👍

@6tZ So why don't you train on the base model? Apparently, it's better for fine-tuning.

Could you please send us your config.json file? Also what are you using to train? AI-ToolKit? Onetrainer? I can't seem to train a single character LORA on any nsfw trained checkpoints. It never works out for me. I just get garbled mishmashes.

same issue.Loras just dont work well with it at all

I trained my character Loars on ZIB, put them to 1.5 and I get great results on all these ZIT models

@paddodust631 I found someone who create a checkpoint that allows you to train a lora on it (https://civitai.com/models/897413?dialog=commentThread&commentId=1117342). I trained it and it worked. However to be honest, i think the quality of the image was worse than ZIB. Maybe i did something wrong idk. Try it out.

The lora should be trained for Z-image-base to work. I was training on z-image-turbo on Ostris, but if trained on ZIB it works, took me a while to figure it out.

how are you guys even getting these to work for z turbo? you just plug the pictures and run it>?