v-pred_v2.0
Carefully Curated Danbooru Dataset (up to September 2025)
Base Model: NoobAI-XL (NAI-XL) V-pred-1.0-Version
When necessary, please use Rescale CFG with values up to around 0.7 (especially for NSFW content).
Quality Tags:
very aesthetic > masterpiece > best quality > good quality > normal quality > worst quality
If you feel like it, you can support me on Ko-fi. Thank you.
Online Generator:https://www.seaart.ai/ja/models/detail/a106ce42854f7b7f65523ad10e03caa8
v-pred_v1.1
Dataset:50k AI-generated images from Aibooru and Pixiv + 10k images from Danbooru
Prompts
people, characters, copyright, style, general tags, rating, masterpiece, best quality, good quality, newest
Negative Prompts
lowres, worst quality, bad quality, bad anatomy, sketch, jpeg artifacts, signature, watermark, old, oldest
Base model:NoobAI-XL (NAI-XL) V-pred-1.0-Version + Illustrious XL 1.0
v-pred_v1.0
Base model:NoobAI-XL (NAI-XL) V-pred-1.0-Version
Use v-prediction and Zero Terminal SNR.
For information on v-prediction and how to use it in ComfyUI, refer to the NOOBAI XL Quick Guide by L_A_X.
As of November 20, checkpoints supporting v-prediction are not available for use in stable-diffusion-webui.
Outside of ComfyUI, you can use the webui's dev branch (likely requiring a .yaml file for each checkpoint) or Forge and reForge.
Additionally, make sure to select "Zero Terminal SNR" under Settings → Sampler parameters → Noise schedule for sampling.
v3.1
Base model:NoobAI-XL (NAI-XL) Epsilon-pred 1.1-Version
v-pred_v0.1
Base model:NoobAI-XL (NAI-XL) V-pred-0.6-Version
v3.0
Base model:NoobAI-XL (NAI-XL) Epsilon-pred 1.0-Version
v2.1
Base model:NoobAI-XL(noobaiXLNAIXL_epsilonPred075 x 0.7 + noobaiXLNAIXL_epsilonPred05Version x 0.3)
v2.0
Base model:NoobAI-XL (NAI-XL) Epsilon-pred 0.5-Version
v1.0
Base model:Illustrious-XL-v0.1
After training, this model merged Lucereon into the U-net.
(0,0.3,0.25,0.2,0.15,0.1,0.15,0.2,0.25,0.3,0.3,0,0.05,0.15,0.25,0.3,0.25,0.15,0.05,0)
License
Description
FAQ
Comments (44)
支持一下 最爱用的一个模型
Nice
Thank you for your work
It's ridden with mosaic censorship. Like the previous one, this series do sfw better by itself than working together with various loras
Can you tell me how to set this up correctly? The images I'm getting are extremely low quality. I'm using reForge.
thank you so much for your work!
The new version has a significant improvement in composition and lighting! However, fingers are still unstable, and the new version always generates extremely close-up pictures, even though I have added "close-up" and "head out of frame" as negative prompts.
Try prompting more of the character's outfit around the lower body and/or reduce the usage of "perspective" and "dynamic pose" in the prompt (you can use complex pose and complex posture for less movement related poses). It could also be the usage of perfect eye loras, but idk what ur prompt is.
good!
Pagman
It seems to be poor at anatomy and fingers.
Good to see a new VP noob.
Sadly, the new version doesn’t have the style of v-pred_v1.1 😭
v2.0 has newer knowledge for new characters, but I still prefer the style of v1.1. I tried training a lora to copy the style, but fingers are even worse ;(
this version needs a fix ! v2 vpred missing arm - missing hand i did only one gen and the first one lost a half arm like a cut trough it wth ? and other issues are listed from other persons ( out of this out of frame thing never got this one maybe he used a prompt that focus on specific body parts) !
v2.0 is great for me
I don't normally comment anywhere, but this time I'm just commenting to say that I have no problems with this version! Until now, I felt that the best version of this checkpoint was vpred1.0, but this new version works great! Thanks for taking the time to create new versions for this checkpoint!👍
Wish EPS ！！ o(￣▽￣)ｄ
ナイス！
Honestly i love this model, it has my favorite style, but awful anatomy makes it almost unusable.
All other most recent models can give you very good looking images even with simplest prompts, this one takes way too much work to get something even decent looking and it's still nowhere close to other models output in terms of anatomy.
Idk maybe it's me doing something wrong, but i just can't seem to make it consistently make anything good.
This model really needs a good anatomy fix and it will be the best model on the site.
what other model do you recommend?
It's a very beautiful model, but unstable—suffering from severe JPEG artifacts, and its rendering of hands and feet is notably worse than other IL-series models. Its adherence to prompts is also relatively poor, especially in responding to artist tags. Despite these numerous issues, its strengths are equally striking: it truly is stunningly beautiful. Your trained models have always been this aesthetically pleasing, and its training dataset is very recent—up to September 2025—making it significantly newer than many other models. This allows it to generate newer characters without needing LoRAs, such as my favorite, "Sameko Saba."
I love it.
Does v2.0 contain IL1.0?
아주 쓸만한 모델 입니다 불완전 하긴 하지만 rtx50 이라면 해결해줄 겁니다
The dataset for this model is up to September 2025, but I cannot use it to generate new characters released before September. Does anyone know how to solve this problem? I clearly saw this character on danbooru, and the data volume should support the generation,For example, it cannot generate reisa, no matter the format is reisa_(blue_archive) or reisa (blue archive)
Also some characters from wuthering_waves can't spawn, but those characters have been around for a long time.
I feel your pain... some characters just don't get picked up by some models for some reason. Best you can do is use a LoRA.
Can try formatting the character like this: reisa \(blue archive\) that sometimes helps. but usually if the two formats you tried don't do it, LoRA's are the way to go.
Because personally trained models cannot crawl hundreds of thousands of images for large-scale training like the IL or Noob teams, their general support for new datasets is definitely not as good as the base IL or Noob models. Therefore, using character LoRAs is still necessary, but it's not entirely without effect. When using this model, for some new characters, you can lower the LoRA weight to avoid LoRA style contamination.
For example, with "almond eye (umamusume)", when using V1.0, I had to set the LoRA weight above 0.85, but it still didn't restore the character features well, and the image style was severely affected by the LoRA. However, in V2.0, although it can't completely avoid using LoRAs, I can reduce the LoRA weight to 0.6 and achieve better results than using 0.9 in V1.0, while the style contamination is also significantly reduced.
Can someone explain why this is smaller than the base model? I would expect it to be larger.
it's fp16 no BF16
Is this checkpoint can use ill lora and noob lora?i see it's base model is noobai,but i see the recommended checkpoint under this checkpoint have ill-xl and noobai(i'm sorry that my English is so poo....
It can use illust LoRa. (At least from my experience, it worked perfectly with illustrious LoRa's)
@Twisowk thx,i use this model version is v2.0,it worked perfectly too,the next step that i want to try the v1.1(i want to know the difference between them XDD)
狗屎Zero Terminal SNR 和 CFG Rescale 会严重影响你其他模型的使用切换模型记得关闭
Insanely nice for me. Please EPS for latest dataset!
v3.0 when
I love this so much
对新角色的支持非常好，远远好于很多宣称支持新角色的模型
Since I discovered this checkpoint, I haven't been able to use anything else. It has a distinct vibe compared to others and produces incredibly emotional illustrations. I mainly create male characters with a NijiJourney-style painting texture, and this model blends them more harmoniously than any other. I’m honestly hooked for life. Thank you so much for this amazing work!
it really is the best in terms of style, no other model even comes close.
it would be the only model I ever used if the dataset for characters etc was updated.
(also it has a slight hand/finger issue but that can be fixed with embeddings/LoRA's)
@junkdoll I totally agree.
come back for anima/rf sdxl and save us
NEED update for illustrious version PLS
Best v-pred checkpoint,bless on you
Details
Files
obsessionIllustrious_vPredV20.safetensors
Mirrors
obsessionIllustrious_vPredV20.safetensors
obsessionIllustrious_vPredV20.safetensors
obsessionIllustrious_vPredV20.safetensors
obsessionIllustrious_vPredV20.safetensors
obsessionIllustrious_vPredV20.safetensors
obsessionIllustrious_vPredV20.safetensors
0b535510n_pr3d_2_0.safetensors
obsessionIllustrious_vPredV20.safetensors
7_obsessionIllustrious_vPredV20.safetensors
Available On (3 platforms)
Same model published on other platforms. May have additional downloads or version variants.