Still, FP8 + Lightning 8-steps Lora is recommended. If you don't like the Fluxy look, use the DPM++ series with Karras (more steps and a higher CFG are required).
End-of-Life
I guess I've learned enough about Qwen-Image, so further testing feels redundant. This repo will not receive any further uploads or attention. I captioned all images and released them as a sign of thanks for testing the beta releases. I hope it helps you, as it was a great resource for me. Good luck!
6,537 captioned images sourced from CivitAI.
Prompts used for generating images (useless for Qwen-Image).
Captions in vulgar and profane style are in the Captions folder.
Caption Example:
This is a digital illustration showing a fucking intense and explicit scene in a movie theater. A muscular dude with short brown hair is sitting behind a blue-haired chick, who's wearing a blue dress and white sandals, and he's fucking her doggy style. His cock is huge and it's ejaculating a fucking lot, with cum dripping down her pussy and onto the seat. He's got one hand over her mouth, and she's looking surprised with wide eyes. The background shows other dudes sitting in red theater seats, looking bored or distracted. The camera angle is straight-on, focusing on the fucking action. The dude's hand is gripping her tight.Prompt Example:
score_9, score_8_up, score_7_up, absurd_res, hi_res, anime_source, (big man sitting on chair in movie_theater, cute girl on lap:1.1), open pants, hug, anal, pussy_juice, stealth_sex, cum,sundress, smug, surprised, exhausted, rolling_eyes, covering_mouth, detailed face, intricate details, hyperdetailed, very aesthetic, motion_lines, <lora:NAI Smooth Boys Style SDXL_LoRA_20r_20e_8i_nr32_a16_Pony Diffusion V6 XL:0.5> <lora:Concept Art Twilight Style SDXL_LoRA_Pony Diffusion V6 XL:0.7>META4 - Helm [Plz read]
Helm needs to be used with META-4 (Strength 0.6) + Helm (Strength 1).
Qwen-Image doesn't respond well to Booru tags.
This is in line with other BETA releases to figure out how to deal with anime (low detail) and realism (high detail). Qwen-Image is not specialized as illustrious for anime, so much of the anime actions need to be done via LoRA. Although, if you're crazy enough, you can do it via prompt only.
META-4 will improve NSFW details to some degree (Still in BETA, not perfect, but better than what Qwen presents).
Helm needs to be used with META-4 (Strength 0.6) + Helm (Strength 1). Merge it with META-4 using your own settings, following the provided code in the article (link in META-4 description).
Trained on 139 randomly picked images (very limited) with no moderation for testing purposes. therefore, it doesn't satisfy an anime enthusiast right away.
7 epochs, 1000 steps, LR 0.0003 (to see if META-4 can act as a refiner).
A dataset with 2 caption variations (Tags, Vulgar) is provided in case you're interested.
If you make one, please ping me.
Datset source: https://civarchive.com/models/1215490/helm-nikke-sdxl-lora-illustrious-or-3-outfits
Last BETA Releases:
META-4
Please read the article related to META-4
https://civarchive.com/articles/18798/qwen-image-nsfw-lora-notes
This version is a linear merge with tuned weights from four releases, each focused on a specific aspect of the training. While it is still far from perfect, it can be useful in some cases.
DO NOT MERGE version 0.4 with the other releases. Overfitting issues. Overfitting occurs when a model learns the training data too well.
v0.4 BETA
I experimented with the learning rate to determine exactly where overfitting will occur.
There is a better skin tone, but signs of overfitting, as well as bad or deformed genitalia, will occur more often than in v0.3.
v0.3 BETA
Experimented with a more friendly and maintainable prompting style. Use one or a combination of them:
Descriptive Style: "A photo-realistic shoot from above featuring a woman in a provocative pose on a bed..."
SDXL Tag-Based Style: "1girl, long hair, breasts, looking at viewer, open mouth."
Segmentation Style:
Sex Acts: Penetration, vaginal intercourse.
Sexual Positions: Missionary position.
Male Genitalia: Large, erect, dark-skinned, circumcised, with visible veins.
This BETA is all about prompting and testing the results. I've removed anime images from the dataset to save time and resources and to speed up the process.
Next, I'll focus on the details and finding a way to eliminate the current issues with anatomy.
v0.2 BETA
Experimented to find a sweet spot for more detailed genitalia
Used Qwen-Image captioning style (This means you need a detailed description of what you want).
The focus was on experimentation rather than quality, hence the BETA.
New auto-generated realistic images were used. Extreme sizes were spotted, but I didn't filter them out.
Pro: Better output compared to BETA 1.
Con: I did a few tests, and writing a wall of prompts is not maintainable. However, Qwen-Image is detail-hungry, otherwise, it takes over, and in the case of NSFW content, we don't want the model's influence.
Next: I'll try mixing Danbooru tags with descriptive captioning, focusing on vulgar slang, and using a better dataset.
v0.1 BETA
This LoRA is primarily trained on Civitai images for experimenting with Qwen-Image LoRA training. 80% of the dataset consists of anime-based images, while the remaining images are semi-realistic, which will likely dominate the output. (mostly vertical sizes)
Using FP8 with 8 steps Lightning LoRA generates acceptable results. All images in the showcase are the best from two batches
Based on the tests I've conducted, the results are promising. This indicates that we don't have the same level of censorship as Flux.
Prompt Guide: I used Joy Caption, Stable Diffusion style of captioning. Example:
[Update: Upon further testing, it turned out that using the SD style for captioning was a bad idea. I will try a different approach in the next beta.]
"""
nsfw, digital painting, close-up, girl with green eyes, black hair in two buns, red halter top, large breasts, hand grabbing her right breast, nipple exposed, gold necklace, light skin, subtle blush, camera angle from below, looking up, soft lighting, realistic style, detailed shading, hand on breast, suggestive, hand touching breast, breast grab, hand on nipple, upper body, focused on face and breasts, red halter top, bouncy hair, soft texture, high detail, hand on nipple, realistic shading, realistic style, soft lighting, subtle blush, looking up, gold necklace, realistic eyes, halter top, realistic breasts, realistic skin, realistic lighting, hand on nipple, detailed shading, high detail, soft texture
"""
Description
meta4 as a semi-refiner
FAQ
Comments (25)
Hey, have you tried sleeping? I hear it's good for you! J/k I just feel like I see an update on this every day. Thanks for all your effort!
lol, well, you can't argue with good advice. Thanks for the advice!
@sweetmax797 Thanks for your work !
Did you find that Qwen is not as flexible as you hoped ?
@jakoc75648 You're welcome!, I'm not sure what you're talking about, but did I say it is not? Flexible in case of what? LoRA? It has nothing to do with flexibility! :D It is a 20B model, meaning you have around 20M trainable parameters, which means the usual way of doing LoRA for SDXL or Flux won't work. So, it is a matter of dataset, captions, hardware, and Qwen-Image can basically do what you ask. It is open source, so I like it! :)
@sweetmax797 Ah dw about it, I definitely mixed things up... Still learning my way around all this, your notes help a lot, and I’m definitely looking forward to seeing more of your work
@jakoc75648 Thank you! Ah, I see. I guess you read the end-of-life as a give-up. If that's the case, not really, the beta is over. I have something in mind, preparing a dataset for it. I'll test it out, and if it works, I'll share the info. else 1 LoRA solution is on the table. The only fun part about all this is during the period of swimming in the unknown. Once you reach the destination, you don't care anymore :) So, have fun in unknown waters! :)
not sure i get it, can't get any good output with this.
well, they were BETA, and very limited. but for the most part is the habits. Qwen can't be used as SDXL or flux. it is very precise. look at the example prompts. it is not just for nsfw, Qwen-Image goes against you if you're not precis.
follow this type of prompt, test on last version or meta4:
Ultra HD, 4K, cinematic composition. (this is like score9. score_8_up ect.)
Example 1: Sex Act: blowjob, woman sucking a penis.
Position: kneeling.
Description: a fully naked woman knees between a nude man's leg and performing blowjob on him. saliva, cum on her face.
------------------
2:
- Sex Acts: vaginal penetration,
- Male Genitalia: erect, large, circumcised, with visible veins and a prominent vein on the head.
- Female Genitalia: aroused, wet, with visible labia and clitoris.
- Breasts: medium, round, with erect nipples.
- Thighs and Buttocks: slim, spread apart.
- Overall Body Types: slim, athletic.
- Vantage Height: eye-level.
- Shot Type: medium close-up.
now this can for qwen image edit, right? how about tongue..
No it 's text-to-image LoRA. tongue, the tongue is out. just use it in you prompt, the tongue will come out :)
@sweetmax797 for tongue kisses, the tongues end up fused and merged.
We need a proper tongue kiss lora for Qwen.
that suck ass more than pony plastic
you're out of your depth.
I'm looking into your dataset, the captions are named like "image_xx.txt" while the images are named "xx.jpeg". They should be named the same with only the extension different. Are you sure your trainings used any caption at all in the process ?
yes, the captioning was done after I decided to close the repo and it was just a nice gesture to ppl who used those tests. so there wasn't any focus on having correct pair names. simple terminal command can fix. in git terminal or ask ai to convert this command to cmd or powershell if ur using windows:
"a=1; for i in {1..20}.jpeg; do mv "$i" "image_$a.jpeg"; a=$((a + 1)); done"
now all jpeg files from 1 to 100 has same name as text files. or if u have ffmpeg installed on your system:
for f in .[jJ][pP][gG] .[jJ][pP][eE][gG]; do [ -f "$f" ] && ffmpeg -i "$f" "${f%.*}.png"; done
this will convert all jpegs to png
and yes I'm sure , if u done it u would no, u cant do it, the training will stop with an error.
@sweetmax797 ok I think you're an alien from outer space. This is the only way you would answer like this xD
Does this work with Qwen-Image-Edit?
yes, Meta4 works with Qwen-Image-Edit (even https://civitai.com/models/1924810/eva-qwen) use a sexy girl image and try this prompt: "a nude man holding her up from behind. she is wearing wet tank top and fully naked lower body. her legs are spread, and the man is holding her legs up. the man's erect penis is in her vagina. full nelson sex style. she has thick thighs and round breasts."
set the model strength to 0.86 or 1.1 depending on the result.
@sweetmax797 hey i tried this prompt wit meta4 but why does he insert in her anus, where i mentioned vagina
@crazy2boy143996 lower the strength to 0.86, play with the strength and higher steps, lots of thing can play a roles why model deviates form the prompt,
does it work with qwen image edit?
yes, META4 version or https://civitai.com/models/1924810/eva-qwen
Plz bear my stupid question, How can I download and install at my ComfyUI? It's zipped file but should I unzip it and place it at the loras folder under models of ComfyUI?
your question is fully reasonable and legit. you need to download META4 https://civitai.com/models/1896397?modelVersionId=2161297 save it in 'loras' folder inside 'models' folder. in your workflow click on 'R' to refresh it and then select the meta4 in lora loader node
”META4 -Helm“Is this version the lora of the anime?