ToriiGate-v0.2

A multimodal LLM finetuned with anime images to be used for captioning of anime images. No limits, no censorship, advanced knowledge yet WIP

Might be very useful for dataset preparation, especially for FLUX/SD3/... models.

Some examples and tricks

You will need any torch >=2.1.2 and dev transformers pip install git+https://github.com/huggingface/transformers to run it.

Example script for single image, for batch processing.

Quantization

To reduce VRAM requirements you can load model in 4 or 8 bit.
Example:

from transformers import BitsAndBytesConfig

nf4_config = BitsAndBytesConfig(
   load_in_4bit=True,
   bnb_4bit_quant_type="nf4",
   bnb_4bit_use_double_quant=True,
   bnb_4bit_compute_dtype=torch.bfloat16
)

model = AutoModelForVision2Seq.from_pretrained(
    model_name_or_path, 
    torch_dtype=torch.bfloat16,
#   _attn_implementation="flash_attention_2", #if installed
    quantization_config=nf4_config,
).to(DEVICE)

Prompt format

For best results stick to one of trained user prompt:

  • Describe the picture in structured json-like format.
  • Give a long and detailed description of the picture.
  • Describe the picture briefly.

First one provides most informative results and should be used when combined with further NLP with other LLM. Second is just a typical large sloppy rubbish output for most vlms. Third gives only meaningful and convenient description and can be used directly as captions.

And do not forget to add system instruction like in provided templates.


Examples below are provided for structured output. Most of them are NSFW.

Examples SFW/Questionable


pic

{
"character_1": "The character in the image is a young girl with short, pink hair and blue eyes. She is depicted in three different poses and expressions. In the top left panel, she is sitting at a table, eating an apple slice with a fork. She has a slightly embarrassed expression with a blush on her cheeks. She is wearing a red and white long-sleeved shirt. In the top right panel, she is shown with her eyes closed and hands together in a prayer-like gesture, smiling widely. In the bottom panel, she is holding a plate with a surprised expression, looking at the viewer.",
"background": "The background is a simple, light beige color with no distinct features. There is a table with a plate of apple slices in the top left panel.",
"texts": "There are some Japanese characters in the image: 'ゴ' in the top left panel and 'ゴ' in the top right panel. There are also some sound effects in the bottom panel, represented by lines and shapes.",
"atmosphere": "The overall atmosphere of the image is light-hearted and cheerful, with the character displaying various emotions from embarrassment to happiness and surprise."
}


pic

{
"character_1": "The first character is a girl with short, multicolored hair that is primarily white with purple highlights. She has animal ears, specifically wolf ears, and a tail that is also white. She is wearing a white sleeveless shirt that is unbuttoned, revealing her cleavage and a black choker with a pendant. She has a playful expression with one eye closed and her mouth open, showing a fang. She is holding a shopping bag in her left hand and her right hand is holding the hand of the second character.",
"character_2": "The second character is a girl with short, white hair and blue eyes. She has pointy ears and is wearing a red shirt with a black design on the sleeve and a black skirt. She is also wearing a black choker. She has a happy expression with her mouth open and is holding hands with the first character. Her right arm is around the first character's shoulder.",
"background": "The background shows a bright, sunny day with a clear blue sky. There are trees and buildings visible, indicating an outdoor setting, possibly a city or town.",
"texts": "The image contains the text 'TOSTANTAN.COM' in the top left corner.",
"atmosphere": "The overall atmosphere of the image is cheerful and playful. The characters appear to be enjoying a sunny day out, possibly shopping or walking together. The bright colors and their happy expressions contribute to a lighthearted and fun mood."
}


pic

{
"character": "The image features a girl with long, grey hair styled in twintails, adorned with a yellow hairclip. She has blue eyes and a slight blush on her cheeks, giving her a cute and approachable appearance. She is wearing a white sleeveless shirt that reveals her bare shoulders, paired with purple shorts that are slightly torn, exposing her thighs. She has a black jacket with green accents draped over her shoulders, and a black bag slung across her body. Her footwear consists of colorful sneakers with green and yellow accents. She is sitting on a ledge, holding a drink in her hand, and looking directly at the viewer with a smile.",
"background": "The background depicts a cityscape with tall buildings and billboards. The sky is clear with a few clouds, and there are some green trees visible, adding a touch of nature to the urban setting. The overall scene is bright and vibrant, with a sense of a sunny day in the city.",
"texts": "None",
"atmosphere": "The atmosphere of the image is lively and cheerful, with a mix of urban and natural elements. The girl's relaxed pose and smile contribute to a friendly and inviting mood."
}


pic

{
"character": "The image features a single female character with long, multicolored hair that is predominantly grey with purple streaks. She has large, expressive yellow eyes and a small ahoge (a single strand of hair sticking up) on top of her head. Her ears are pointy, and she has prominent, curved black horns with purple stripes. She is wearing a white shirt that is unbuttoned, revealing her bare chest and navel. The shirt is partially see-through, hinting at her lack of a bra. She also has a black choker with a purple gem in the center around her neck. Her expression is playful, with her tongue slightly sticking out and a blush on her cheeks. She is looking directly at the viewer, creating a sense of engagement.",
"background": "The background is blurry, with soft, pastel colors that suggest a dreamy or ethereal atmosphere. There are no distinct objects or features, keeping the focus on the character.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is playful and slightly provocative, with the character's revealing outfit and flirtatious expression. The soft, blurry background adds a dreamy and intimate feel to the scene."
}


pic

{
"character": "The image features a single female character with long, blonde hair that reaches down to her waist. She has blue eyes and a blush on her cheeks, giving her a shy or embarrassed expression. Her mouth is closed, and she is looking directly at the viewer. She is wearing a black shirt that is open, revealing her bare chest and navel. The shirt is being held open by her hands, which are positioned on either side of her chest. She is bottomless, with a small bandaid placed strategically over her genital area. Her body is depicted in a realistic style, with detailed shading and highlights.",
"background": "The background is simple and white, with no additional elements or objects present.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is intimate and slightly provocative, with the character's shy expression and revealing pose creating a sense of vulnerability and allure."
}


pic

{
"character": "The image features a girl with long brown hair styled in double buns, adorned with a golden hairband and a hair ornament resembling a ship's superstructure. She has grey eyes and a slight blush on her cheeks, giving her a shy yet inviting expression. She is wearing a white towel that wraps tightly around her body, accentuating her large breasts and curvy figure. The towel is held in place by her hands, one of which is positioned near her chest, while the other is slightly raised. She has a closed mouth and a subtle smile, and her overall demeanor is soft and inviting.",
"background": "The background is blurry, suggesting an indoor setting with soft lighting. There are indistinct shapes that could be furniture or other objects, but they are not clearly defined.",
"texts": "There is a speech bubble with a musical note, indicating that she might be humming or singing.",
"atmosphere": "The atmosphere of the image is warm and intimate, with a hint of playfulness. The girl's expression and posture suggest a sense of comfort and relaxation, creating a cozy and inviting mood."
}


pic

{
"character": "The image features a single female character with long, flowing grey hair adorned with a white flower and a maid headdress. She has heterochromia, with one eye being blue and the other yellow. Her expression is neutral, with parted lips and a slight blush on her cheeks. She has cat ears on top of her head, which are white with pink inner fur, and a grey cat tail that is positioned between her legs, partially covering her crotch. She is completely nude, revealing her medium-sized breasts with visible nipples, and her navel. Her legs are spread apart, and she is lying on her back, with one hand resting on her stomach and the other raised near her head.",
"background": "The background is a simple, solid grey color, which makes the character stand out prominently.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is sensual and intimate, with a focus on the character's nude form and cat-like features. The neutral expression and blush add a touch of innocence to the otherwise provocative pose."
}


pic

{
"character_1": "The character is a girl with orange hair styled in a ponytail, adorned with a pink hair ornament. She has green eyes and is wearing sunglasses on her head. Her expression is one of pleasure, with her tongue out and a blush on her cheeks. She is wearing a green tank top that is lifted up, exposing her large breasts with visible nipples. She also has a necklace with a heart pendant and earrings. Her body is glistening with sweat, and she is holding the hand of another person, guiding it to her breast.",
"character_2": "The second character is not fully visible, but their hand is seen guiding the girl's breast. The hand is holding the girl's breast firmly, and there is a sense of intimacy and connection between the two characters.",
"background": "The background is blurry, with some indistinct shapes and colors, suggesting an indoor setting with curtains and a window. The focus is primarily on the characters in the foreground.",
"texts": "There are several speech bubbles and sound effects in Japanese, indicating moans and sounds of pleasure. The text includes '...♡', 'あっ', and 'はっ'.",
"atmosphere": "The atmosphere is intimate and erotic, with a strong sense of physical and emotional connection between the characters. The girl's expression and body language convey pleasure and arousal, while the presence of sweat adds to the intensity of the moment."
}


Most of outputs for such images are nice as is, even more explicit should be fine.

NSFW and complex images

However, when you moving to a more complex scenes, it might start to make mistakes.

You should add booru tags (even from wd3tagger ensemble is fine), it will improve accuracy significantly!

1
2
3
tags='1girl, standing, looking at viewer, ...'
user_prompt+=' Also here are booru tags for better understanding of the picture, you can use them as reference.'
user_prompt+=f' <tags>\n{tags}\n</tags>'

Examples for complex/NSFW with tag guidance


pic

{
"character_1": "The image features a girl with long brown hair and brown eyes, who is straddling a male figure in a cowgirl position. She has a flushed face, with a blush spreading across her cheeks and nose, and her eyes are half-closed, giving her a dazed and aroused expression. She is wearing a white collared shirt that is open, revealing her large breasts, which are exposed and glistening with sweat. Her nipples are visible and erect. She is also wearing a black bra that is pulled down, a pair of pink panties with a bow, and brown pantyhose. Her legs are spread wide, and she is leaning forward slightly, with one hand touching her mouth and the other resting on the male figure's stomach. She has earrings and a grin on her face, indicating pleasure.",
"character_2": "The male figure is not fully visible, but his presence is implied by the girl's position and the fact that he is lying on his back. His hands are not visible, but his stomach is exposed, and the girl is resting one of her hands on it.",
"background": "The background is blurry, but it appears to be an indoor setting with curtains and a ceiling with recessed lighting. The lighting is warm and creates a soft, intimate atmosphere.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is intimate and erotic, with a focus on the girl's expression and body language, which convey pleasure and arousal. The warm lighting and blurred background add to the sense of privacy and intimacy in the scene."
}


pic

{
"character_1": "The first character is a woman with black hair styled in double buns, adorned with bun covers. She is wearing a traditional Chinese outfit, consisting of a blue and red dress with a white belt and detached sleeves. She has a bracelet on her wrist and is holding a piece of green leafy vegetable in her hand. Her expression is focused, and she appears to be looking at the second character.",
"character_2": "The second character is a woman with black hair styled in twin drills, wearing a revealing crop top and a dudou. She is bound with ropes around her arms and legs, and her breasts are exposed with cum dripping on them. Her body is covered in cum, with visible cum in her pussy, ass, and mouth. There are tally marks written on her body, and she is kneeling with her legs spread wide. She has a distressed expression, with tears in her eyes and cum dripping from her mouth.",
"background": "The background features a dimly lit setting with a large, glowing moon or light source behind the first character. There are some structures and a railing visible, suggesting an outdoor or rooftop setting.",
"texts": "None",
"atmosphere": "The atmosphere is intense and explicit, with a strong focus on BDSM and sexual themes. The scene is graphic and provocative, highlighting themes of bondage, dominance, and submission."
}


pic

{
"character_1": "The character is a girl with long blue hair and pink eyes, wearing a blue hat with a feather. She has large breasts and is depicted nude, with her legs up and restrained by cuffs. Her nipples are visible, and she has heart-shaped pupils. She is wearing black pantyhose that are torn, revealing her pubic hair and vagina. Her mouth is open, and she appears to be in a state of arousal.",
"character_2": "The second character is a boy, though only his penis is visible in the image. His penis is ejaculating inside the girl's vagina, which is shown in a cross-section view. The penis is censored with mosaic censoring.",
"background": "The background is minimal, with a focus on the characters. The setting appears to be a bed with white sheets.",
"texts": "None",
"atmosphere": "The atmosphere is highly sexual and explicit, with a focus on the intimate and graphic nature of the scene. The girl's expression and the heart symbols around her head suggest pleasure and arousal."
}


pic

{
"character_1": "The character in the image is a girl with long, white hair and purple eyes. She is wearing a black bra and black panties, which are partially visible. Her large breasts are prominently displayed, and she has a flushed expression on her face. She is barefoot, with her toes and sharp toenails visible. She is holding a black object, presumably a penis, with her toes, and her legs are spread apart, revealing her anus and the cleft of her vulva. Her body is glistening with sweat, and she has a slightly open mouth, suggesting a state of arousal or exertion.",
"background": "The background is not very detailed, but it appears to be an indoor setting with a bed or couch where the girl is lying. The lighting is dim, creating a somewhat intimate and private atmosphere.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is highly sexual and explicit, focusing on the girl's provocative pose and the intimate act she is performing. The sweat on her body and her flushed expression add to the sense of arousal and intensity of the scene."
}


pic

{
"character_1": "The first character is a girl with long, light brown hair and green eyes. She is nude, with large breasts and visible nipples. She is positioned on top of the second character, her body pressed against the other girl's. Her expression is one of pleasure, with her mouth open and eyes closed. She is wearing white gloves and black socks. Her hands are gripping the second character's ankles, pulling them up.",
"character_2": "The second character is a girl with long, light brown hair and closed eyes. She is also nude, with large breasts and visible nipples. She is lying on her back with her legs spread apart and her hands clenched into fists. She is wearing white thigh-highs and white gloves. Her expression is one of pleasure, with a blush on her cheeks.",
"background": "The background is a simple, light-colored room with a pink bedspread. There is a cross-section view of the internal cumshot, showing the penetration and ejaculation inside the second character's vagina.",
"texts": "None",
"atmosphere": "The atmosphere is highly sexual and intimate, with both characters engaged in a passionate and explicit act. The scene is intense and focused on the physical pleasure and connection between the two characters."
}


pic

{
"character": "The image features a single female character, depicted in a dogeza position, which is a traditional Japanese form of apology or bowing. She is completely nude, with her long, dark hair cascading down to the floor. Her body is covered in sweat, indicating nervousness or physical exertion. Her hands are pressed against the floor, and her head is bowed deeply, showing a sense of deep apology or submission.",
"background": "The background consists of a red, patterned floor with a geometric design. There is a table with a lit lamp on the left side of the image. On the right side, there is a neatly folded purple bodysuit and a pair of boots, indicating that the character has removed her clothing before assuming the dogeza position.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is one of intense submission and apology. The character's nude form and the sweat on her body emphasize her vulnerability and the gravity of her actions. The red and orange tones of the background add to the dramatic and intense mood of the scene."
}


pic

{
"character_1": "The central character is a girl with orange hair styled in a braid and adorned with a feather hair ornament. She has purple eyes and is wearing a dirndl dress, which is partially open to reveal her breasts. Her nipples are erect, and she is covered in cum, with some of it on her hair, body, and mouth. She is engaged in a gangbang with multiple boys, performing oral sex on one while being penetrated vaginally and anally by others. Her legs are spread wide, and she is holding a penis in her hand, giving a handjob.",
"character_2": "There are four boys in the image, each with visible penises. One is being sucked by the girl, another is being held by her hand, and the other two are penetrating her vaginally and anally. Their faces are not visible, focusing the attention on the girl.",
"background": "The background is simple and unadorned, keeping the focus on the characters and their actions.",
"texts": "There are small heart symbols near the girl's head, indicating pleasure or affection.",
"atmosphere": "The atmosphere is highly sexual and explicit, with a focus on group sex and the girl's multiple sexual activities. The image is graphic and intense, emphasizing the physical and sexual interactions between the characters."
}


pic

{
"character": "The image features a female android with a humanoid form. She has a sleek, white, and metallic body with blue and purple accents. Her eyes are glowing blue with blue sclera, and she has a slight blush on her cheeks. Her lips are parted, and she has a small smile on her face. She is looking back over her shoulder with a contrapposto pose, emphasizing her large breasts and curvaceous figure. Her skin is white, and she has a robotic appearance with visible joints and mechanical parts. She is completely nude, with multiple used condoms hanging from her body, including one left inside her vagina. There is cum dripping from her anus and vagina, and her ass is prominently displayed in the foreground. She is holding a used condom in her right hand, and her left hand is resting on her hip.",
"background": "The background is a gradient of dark blue to black, with no distinct features or objects, keeping the focus on the character.",
"texts": "None",
"atmosphere": "The atmosphere of the image is explicit and sexual, with a focus on the aftermath of a sexual encounter. The character's expression and pose suggest a sense of satisfaction and confidence."
}


pic

{
"character_1": "A blonde-haired girl with cat ears and a tail, positioned on all fours. She has a tattoo on her lower back and is wearing a heart-shaped earring. Her eyes are closed, and she is engaged in oral sex, with cum visible on her face and in her mouth. Her breasts are exposed, and she has a blush on her cheeks.",
"character_2": "A dark-skinned girl with white hair and dragon horns, positioned on the right. She has large breasts, which are being sucked by the third girl. She is wearing glasses and has a tattoo on her arm. Her eyes are open, and she is looking at the third girl. She is also sweating and has cum on her body.",
"character_3": "A girl with black hair and glasses, positioned between the other two girls. She is sucking on the dragon girl's breast and has her hand on the blonde girl's ass. She is wearing earrings and has a tattoo on her arm. Her eyes are closed, and she is also sweating.",
"background": "The background shows a cozy room with a couch and a table with books. There is a moogle sitting on the couch, watching the scene.",
"texts": "None",
"atmosphere": "The atmosphere is intense and erotic, with all three girls engaged in sexual activities. The scene is explicit, with visible cum, sweat, and tattoos. The overall mood is one of passion and intimacy."
}


pic

{
"character_1": "The first character is a girl with long black hair, wearing a white headscarf and a brown scarf. She is depicted in multiple scenes, engaging in sexual activities. In one scene, she is kissing a boy passionately, with her hands on his face. In another scene, she is seen having sex with the boy, with her penis penetrating him. She is also depicted with ahegao expression, indicating pleasure.",
"character_2": "The second character is a boy with short black hair, wearing a brown hat and a grey outfit. He is depicted in multiple scenes, engaging in sexual activities with the girl. In one scene, he is kissing the girl passionately. In another scene, he is seen having sex with the girl, with her penis penetrating him. He is also depicted with an erection and sweat on his body.",
"character_3": "The third character is a girl with long black hair, wearing a white headscarf and a brown scarf. She is depicted in multiple scenes, engaging in sexual activities. In one scene, she is kissing the boy passionately, with her hands on his face. In another scene, she is seen having sex with the boy, with her penis penetrating him. She is also depicted with ahegao expression, indicating pleasure.",
"background": "The background is plain grey, with no distinct features or objects.",
"texts": "None",
"atmosphere": "The overall atmosphere of the image is explicit and sexual, with a focus on the intimate and intense interactions between the characters. The scenes depict a threesome with a futanari element, where the girls have penises and are engaging in double anal penetration and double penetration with the boy. The characters are shown with expressions of pleasure and sweat, indicating the intensity of the activities."
}


pic

{
"character": "The image features a single female character with long brown hair styled in twintails. She is lying on her back with her legs spread wide apart. Her eyes are empty, and her mouth is open, giving an expression of exhaustion or pleasure. Her body is covered in cum, with a significant amount pooling around her vagina and dripping from her pussy. There is also cum on her stomach and breasts. Her nipples are visible, and her medium-sized breasts are exposed. There is body writing on her stomach, partially obscured by cum. She has pale skin and visible pubic hair.",
"background": "The background is simple and white, with a vignette effect around the edges of the image. There is steam or mist around the character, adding to the atmosphere.",
"texts": "There are several heart symbols and 'uh...' sounds around the character, indicating sounds of pleasure or exhaustion. The 'Tally' symbol is also visible on the character's stomach.",
"atmosphere": "The overall atmosphere of the image is explicit and intense, focusing on the aftermath of a sexual encounter. The character's expression and the cum on her body emphasize the raw and graphic nature of the scene."
}

Still some small mistakes can occur

Something more complex?

No way at this moment, unless you are willing to correct it yourself. Both SFW/NSFW where a multiple character with complex interaction is present, or something very unusual happens have a low accuracy.

Here are examples where if fails


pic

{
"character_1": "The first character on the left is a girl with blonde hair tied in a ponytail. She is wearing a dark-colored kimono with a red sash and sandals. She has a surprised expression on her face and is looking at the girl next to her.",
"character_2": "The second character is a girl with blonde hair and green eyes. She is wearing a white kimono with a red sash and a hat with a frog ornament. She is holding a sparkler and smiling.",
"character_3": "The third character is a girl with long green hair and green eyes. She is wearing a blue kimono with a white pattern and sandals. She is holding a bucket and pouring water on the sparkler held by the second character.",
"character_4": "The fourth character is a girl with blue hair and blue eyes. She is wearing a blue kimono with a white pattern and sandals. She is holding a sparkler and has a surprised expression on her face. She also has ice shards protruding from her back.",
"background": "The background features a traditional Japanese setting with a wooden fence, a stone lantern, and some greenery. The ground is paved with stone tiles, and there is a small patch of grass.",
"texts": "None",
"atmosphere": "The atmosphere is festive and playful, with the characters enjoying a summer festival. The scene is lively and filled with the excitement of playing with sparklers."
}


pic

{
"character_1": "A large, monstrous creature with a single eye and multiple tentacles. It has a rough, brownish skin and sharp teeth. The creature is engaging in sexual activity with a cat girl, who is positioned on top of it. The cat girl has black hair, cat ears, and a tail. She is nude, with her breasts exposed and her legs spread wide. The tentacles are penetrating her vagina and wrapping around her body.",
"character_2": "A green-skinned lizard-like creature with a large, muscular build. It has a reptilian head with a snout and sharp teeth. The creature is engaging in sexual activity with a blonde-haired girl with rabbit ears and a tail. The girl is nude, with her breasts exposed and her legs spread wide. The lizard is penetrating her vagina with its penis.",
"character_3": "A dark-skinned girl with short black hair and cat ears. She is nude, with her breasts exposed and her legs spread wide. She is positioned on all fours, engaging in oral sex with the large, monstrous creature. Her tail is visible, positioned behind her.",
"character_4": "A red-haired girl with fox ears and a tail. She is nude, with her breasts exposed and her legs spread wide. She is positioned on all fours, engaging in oral sex with the green-skinned lizard-like creature. Her tail is visible, positioned behind her.",
"character_5": "A brown-haired girl with fox ears and a tail. She is nude, with her breasts exposed and her legs spread wide. She is positioned on all fours, engaging in oral sex with the large, monstrous creature. Her tail is visible, positioned behind her.",
"background": "The background is dark and shadowy, with a stone floor and some pillars. The setting appears to be a dungeon or a dark, underground chamber.",
"texts": "None",
"atmosphere": "The atmosphere is dark and intense, with a strong focus on the sexual activities between the characters. The scene is explicit and graphic, with a sense of dominance and submission among the characters. The presence of the monstrous creatures adds an element of danger and non-human interaction."
}


pic

{
"character_1": "The first girl has long, purple hair and is wearing a school uniform with a pleated skirt. She is positioned on her back, with her legs spread apart. Her face is flushed, and she is looking at the viewer with a slightly open mouth. She is engaged in a sexual act, with a penis penetrating her vagina. Her breasts are visible, and there is cum on her body and in her pussy.",
"character_2": "The second girl has short, white hair and is also wearing a school uniform. She is positioned on her knees, leaning over the first girl. Her face is close to the first girl's pussy, and she is performing oral sex. Her hands are on the first girl's thighs, spreading them apart. There is cum on her face and hands.",
"character_3": "The third girl has short, red hair and is wearing a school uniform. She is positioned on her back, with her legs spread apart. Her breasts are visible, and she is looking at the viewer with a flushed face. She is engaged in a sexual act, with a penis penetrating her vagina. There is cum on her body and in her pussy.",
"background": "The background shows a bedroom with a bed and a nightstand. There is a basketball visible in the corner of the room. The lighting is dim, with a blue hue illuminating the scene.",
"texts": "There is a speech bubble with an image of the first girl's ass being penetrated, showing cum dripping out.",
"atmosphere": "The atmosphere is highly sexual and explicit, with all three girls engaged in a threesome. The scene is intense and intimate, with a focus on the sexual acts and the physical interactions between the characters."
}

Also sometimes in can loops. Use sampling to avoid it, but still LLM correction/checking if it is in agreement with booru tags, or manual checks needed

Edit
Pub: 28 Sep 2024 19:42 UTC
Edit: 29 Sep 2024 10:06 UTC
Views: 98