This model is build for all anime lovers.
I continued training this model on a 1M NSFW anime dataset. Training is still ongoing, but it's already functional—enjoy!
Training is far from finished, with only about 10% completed so far.
Important
Prompts are absolutely crucial. If your prompt is too simple, the results will look just like basic anime model outputs. But if you include highly detailed scene descriptions, the outcome will be on a completely different level.
Prompting Tip: A blend of rich natural language (environment, action, character, style) paired with strategic prompt tags yields the best results; overly minimal prompts often lead to overly simplistic images.
An anime-style digital illustration in a dark fantasy gothic aesthetic with a dramatic low-angle perspective that emphasizes depth and scale, featuring a demon girl with long flowing white hair and large curved black horns seated on an elevated stone ledge; she has striking heterochromia—one glowing red eye and one cyan blue eye—along with pointed elf ears and red facial markings, looking down at the viewer with a mischievous smile; she wears an intricate black lace outfit with sheer lace stocking tights, dark red gloves, and shiny black Mary Jane heels with blue jewel details; in her gloved hand she holds an ornate gothic staff topped with a glowing red orb crackling with violet lightning trails; the backdrop showcases a massive, detailed full moon illuminating the dark night sky, framed by gothic castle spires, arching stone architecture, floating pink petals, and a dramatic foreshortened composition that enhances the immersive visual impact.Recommended Settings:
Resolution: 1024–2048px
Steps: 3-8
CFG Scale: 1
Sampler: Euler, Euler A / DPM++ SDE (or
er_sde)Scheduler: Simple, Normal
Supported Quantizations / Precision:
BF16, FP8, INT8, and INT4 formats are all supported!
V1.0
The first stable release. It performs exceptionally well across various art styles and already shows the potential to rival models like Pony in NSFW content. It works great with natural language prompting and understands most mainstream characters, though artist tags are not supported. Training progress is at roughly 5%.
V1.5
A recent checkpoint with impressive performance. It excels in multi-character scenes and character details while mastering a wider variety of art styles. I've made it my go-to anime model—it outperforms models like Pony and IL. It understands most mainstream characters, but artist tags are barely supported. Training progress is at roughly 10%.
Description
FAQ
Comments (34)
这得什么显卡才能跑啊，
I'd recommend the INT4 version. It only needs 6GB VRAM, so pretty much any GPU can handle it.
It just keeps getting better and better. This model pays more attention to your prompts than most other Krea2 models, and it's really very able with a variety of media.
This is fantastic for western cartoons too and the smallest version is really fast with 4-step lora.
maybe we need to know what the range of "most mainstream characters" is
For example, characters from some well-known anime-style games
@civitadmin Could you elaborate on the specific scope? Something like a Danbooru database with some cutoff date for how recent it goes? Or, to put it another way, which ones would you consider well-known anime-style games? Genshin Impact? Wuthering Waves? Arknights? Blue Archive? Is there a defined character set?
@Aetherlyn Well, my dataset has data up to August 2026, but the model still hasn't picked up the new characters. It does get better at the characters it already knew, though honestly, I have no clue how many it actually knows in total.
Wow, v1.5 is really good at anime style and screencaps!
does the dataset have ai generated images? it also looks like those more than v1
76/5000
I would like to know if this model, when compared to the original model, will result in any loss of image details. The original model can generate complex backgrounds. Will this model be able to maintain this feature? Of course, I will also try it myself. Thank you, author!
To be honest, the background isn't as good as the original model's. This is mostly because the training images had very simple backgrounds.
@civitadmin Thank you for your reply! May I ask if this deficiency will be addressed in the upcoming training sessions?
@JIANX yes
I am still exploring how to fully leverage the model's capabilities. Currently, I find that more complex and comprehensive prompts consistently yield significantly better outcomes than shorter, simpler ones. Prompt like this can create complex backgrounds
A cinematic masterpiece photo of a young woman standing in a modern luxury bathroom.
Subject & Pose:
She turns slightly with an S-curve silhouette, one hand gently resting on her waist and the other hanging down naturally. Her posture is relaxed and confident. She tilts her head slightly down, eyes gazing aside with a soft, dreamy, and bedroom-eyed expression, creating a lazy, private, and intimate atmosphere.
Fashion & Appearance:
She wears a light gray, short-sleeved, tight-fitting knit top and a beige lace side-slit hip-hugging pencil skirt, showcasing elegant and sexy curves. On her feet are striking black patent leather high heels with iconic red soles, creating a bold contrast. She is wearing sheer transparent stockings that accentuate her slender, well-proportioned legs. Her reddish-brown hair cascades smoothly over her shoulders. She features defined facial contours, exquisite yet subtle natural makeup, soft lip color, and fair, delicate skin.
Foreground:
A white fluffy plush carpet sits in the foreground with her high heels resting on it; its slightly curled edges add a touch of realism. Next to it is a washbasin with a sleek, polished dark marble countertop reflecting ambient light and shadows. Beneath the counter, an open storage compartment holds neatly stacked, soft white towels and a richly detailed woven storage box, adding a warm, lived-in touch.
Midground & Background:
The female subject dominates the center-right midground. The background features expansive dark marble walls and floors with natural textures and a glossy finish reflecting soft light. A vertical warm yellow light tube mounted on the wall emits a soft halo, illuminating the space. In the distance, a glass partition reveals a blurry shower area with a faint metal frame, enhancing depth of field.
Composition & Lighting:
Low-angle shot, looking upward to emphasize her height and long leg lines, giving the architectural background a layered, structured depth. The visual guidance flows naturally from the carpet and heels up her legs, torso, and finally to her face. The mood is defined by warm yellow side-lighting from the light tube, casting soft shadows that contour her three-dimensional figure and creating delicate reflections on the marble surfaces.
Aesthetic Tone:
A cool color palette perfectly balanced by localized warm lighting. It captures a candid, unpretentious, and accidental moment in daily life, embodying the independence and elegance of a modern urban woman. Extremely detailed, high-end photography, rich texture, cinematic storytelling, 8k resolution, highly realistic, depth of field.
We hope to remove all special effects and fluids during sex scenes, as this is not conducive to i2v video
get
In your introduction, you mentioned, “It understands most mainstream characters.” Does that mean this fine-tuned model recognizes more anime characters that the original official model didn’t recognize?
yes
@civitadmin Thanks for your reply! I have another question: I noticed that the filename of the int4 model is int4_w4a4. I’d like to confirm whether this model is regular int4 or int4_convrot.
@nguspring int4_convrot
@civitadmin I still can't draw most of the game and anime characters I want, such as Exusiai from Arknights. Is it a problem with the character trigger word, or is it because it's not in the training set? I'm using "exusiai from arknights" as the character trigger word. I hope the new version will be released soon.
I just want to congratulate you for this project, its only 10% trained but the results are astonishing! Krea 2 is a great base model for anime it seems
Yes, I am still exploring how to fully leverage the model's capabilities. Currently, I find that more complex and comprehensive prompts consistently yield significantly better outcomes than shorter, simpler ones
Prompts are absolutely crucial. If your prompt is too simple, the results will look just like basic anime model outputs. But if you include highly detailed scene descriptions, the outcome will be on a completely different level.
If you want to draw clean 2D picture, starting with this prompt is highly effective:
masterpiece, best quality, anime_style screencap, anime screenshot style, manga style, 2D anime style, no highlights, flat colors, clean lineart, simple shadingThank you so much for your dedication. If what you said is 100% complete, it would be a remarkable feat!
Hi! Thanks for sharing this great model. Since this is a Turbo version, would it be possible for you to release the base (non-turbo) checkpoint for training LoRAs on top of it？ Thanks!
Thanks for the suggestion. I'll open-source the base weights in the next version.
This is a brilliant base model: bright, clean and packed with details. Its sharpness is unmatched by any alternative. Thanks for sharing, and I look forward to future iterations.
this is amazing, when this model gonna have more then 10% ? stuff like futas etc are still not that great, i hope 2.0 v is soon, and thanks for your work ❤️❤️❤️
My tests are already producing futa concepts, but the prompting needs to be really precise.
The Best Krea 2 Model 😍.
I just want to note that this model, as high detail and aesthetic as is, is currently VERY "portrait" heavy. It's missing actions and is overriding Krea2's understanding of "action."
Feel free to demo this for yourself: "actively traversing the ridge of the unstable stone wall, shoes upon the narrow top surface with a dynamic, balanced gait. One hand is held out slightly for stability, her body tilted as she steps forward". Krea2 gets this every time. Krea2 Anime struggles.
Thanks for the efforts so far - keep it up!