AstroNSFW V2 - ZIG (Z-Image Base Edition)
The Evolution of Anatomy. V2 is a complete rebuild from the ground up, switching from tag-based training to high-fidelity Natural Language (Full Text) Captioning.
✨ What's New in V2?
This isn't just an update; it's a completely new training approach.
dataset: 381 hand-picked, high-resolution 4K images (no automated bulk-scraping).
Captioning: Switched to Full English Sentences. The model now understands context, lighting, and complex interactions better than simple tags.
Diversity: Massive expansion in ethnicities, ages, and body types.
Lighting Mastery: Trained on everything from harsh studio flash and neon lights to soft dappled forest sunlight.
Flexibility: Balanced dataset between indoor/outdoor settings and a wide variety of camera angles (macro, wide, drone, POV).
📖 How to use V2 (Important!)
Since this version uses Natural Language, you don't need to spam tags anymore. Describe what you want to see in a sentence.
Trigger Word:
AstroNSFW-V02-ZIGRecommended Weight:
0.8-1.0Prompt Style: "A high resolution photo of [TriggerWord], a woman standing in..."
Example Prompt:
AstroNSFW-V02-ZIG, A high-resolution photograph of a nude woman with dark skin standing in a rainforest waterfall, dappled sunlight on her skin, wet texture, athletic build.
📋 Comparison: V2 vs. V1
Feature V2 (New):
Prompting Natural Language / SentencesTag / Booru Style
Anatomy High-Fidelity / Texture FocusGeneral Shapes
Dataset 381 Selected High-ResGeneral Dataset
Understanding Context & Lighting awareKeyword dependent
📦 Legacy Info: V1 Tags
(For users still preferring the V1 Tag-style workflow)
Click to view V1 Tags
white panties, white background, wet, water, urethra, upper body, undressing, underwear, uncensored, twintails, towel on head, topless, thighs, thighhighs, tanlines, tan, sweat, striped, squatting, spread pussy, spread legs, solo, smile, small breasts, skirt lift, skirt, sitting, simple background, short hair, shirt lift, shirt, saliva, rock, ring, red eyes, realistic, pussy juice, pussy, pubic hair, pool, ponytail, plaid shirt, plaid, pink shirt, pink nails, piercing, photo background, penis, pectoral focus, panties, outstretched arms, outdoors, open shirt, open mouth, open kimono, open clothes, on stomach, on back, nude, no panties, no humans, no bra, nipples, necklace, navel piercing, navel, nail polish, mole on breast, mole on ass, mole, medium breasts, masturbation, male pubic hair, male focus, lying, lower body, looking back, looking at viewer, long hair, lips, large breasts, lamia, kneeling, jewelry, indoors, holding hands, hetero, head out of frame, grabbing, from behind, freckles, flat chest, fingernails, fingering, female pubic hair, female masturbation, feet, eye focus, earrings, dress, depth of field, day, dark-skinned female, dark skin, corset, completely nude, collarbone, clothes lift, closed eyes, close-up, clitoris piercing, clitoris, chest hair, brown hair, brown eyes, breasts apart, breasts, breast focus, bracelet, bra, bottomless, blurry background, blurry, blue nails, blue background, blonde hair, black thighhighs, black hair, bent over, bed, beach, barefoot, astronsfw-v01-zig, ass focus, ass, arms behind back, armpits, anus, animal, all fours, 1girl
💬 Feedback
I spent a lot of time hand-picking the images for V2 to fix the anatomy mutations from previous versions. Please leave a review with your generations!
Description
381 Handpicked images
Full englisch text prompting
Wide range of ethnics
wide range of lighting
indoor vs outdoor balance
wide range of camera angles and poses
FAQ
Comments (22)
Plan for V3:
- Fixing Nipples (more depth)
- Fixing Vagina Lipps (more depth and details)
- Fixing Buthole (more depth and details)
- ADDING:
1. more night images
2. more lingerie clothing
3. more pose images
4. more multiple person nsfw images
5. a few mroe pubic hair images
6. breast areola size stearing
i think z-image areolas are too big. I've seen lots and z-image ones usually come from women with super huge breasts. not to forget, the lack up hard/puffy nipples.
@Melodic_Possible_582589 I Am working on it alread, I noticed it also. I let you know when my next version is out there.
@Melodic_Possible_582589 My impression is exactly the opposite. Everyone has nipples and areolas that are as small as Belgium.
@heinerheinervonvielen417 Let me google Belgium! xD . But yeah i am working on it, V3 is already running with a extendet dataset and fixed issues. Wish me luck it dosent crash again XD
@Astroburner That was referring to the standard model. Unfortunately, I haven't tried your Lora yet. It would be great if you could perhaps also take curvier body shapes and breast shapes into account in your next training rounds. That would be really fantastic. Like version O of this Lora, for example. https://civitai.com/models/2318875?modelVersionId=2647060
@heinerheinervonvielen417 sure thing V3 is already running but in v4 i wil ladd around 100 - 150 new datasets with more ethnics and body types so curvy it is xD. TBD
@Astroburner honestly what I would love and even pay for is a decent checkpoint that uses a larger set of regular looking people. It feels impossible to get regular looking faces and bodies. Even the curvy women usually look like super models with perfect skin etc
@SaburoDio True story!
@SaburoDio I totaly understand you, and I am working already to collect the perfect images. My standard is quit high, nothing under 2k Resolution. xD I am on it. Still need to test a few settings to addapt more body control and body detail control in the lora
@Astroburner appreciate that my dude! Imo ZIT is already a great model to use and I am curious what the future will bring
i will be testing this on webui neo when it comes out. I know comfy produces better images, but it is what it is.
Let me know the results, maybe I can adapt a few settings in my next lora training
6N7J6HQHQ9DWBK1CZ48W00JFT0
Seriously?
? What do you mean?^^
@Astroburner this is name of downloaded file.
@210881175 ups, ok, maybe i missed this one xD
lora key not loaded: lora_unet_output........
I dont knoiw why but on Draw Things is interpreting this as SDXL lora...?
Select the base model manually, the auto select feature when importing is awful.
V3 for Z-Image Turbo was a 50/50 result. changing now to Z-Image Base. my other loras was absolut great with it. takes some time to cook^^
please too lora for waillustriousSDXL(IL-too realistic lora please)



















