Noct Q
Qwen Image 2.1, uncensored out of the box, packaged for ComfyUI in two versions: Turbo (6 steps) and Base (full steps). Nudes and explicit scenes render without a LoRA.
What's new in V3
Turbo rebuilt and sampled in 6 steps: sharper images and seeds as varied as the full model
In a blind test on the same 150 prompts and seeds as V2 Turbo, fewer defects with the same prompt adherence and explicit rendering
Base is a new model, stronger on explicit scenes
Settings V3
Turbo: 6 steps, euler, no CFG, 832x1248.
Base: 25–50 steps, euler, simple. V3 Base: cfg 3 with the negative prompt, both set in the workflow.
Prompt examples
The prompts of three V3 Turbo showcase images, written like captions of the finished picture: the medium first, then the woman, what she does, where, the light and the framing. All three use V3.0 Turbo int8 convrot in the V3 workflow (6 steps) at 832x1248, and the same seed should give the same picture.
Attic room, seed 96060
An amateur photograph of a nude woman in her twenties with long light-brown hair and full hips, kneeling on an unmade bed in a Parisian attic room with two round windows, stretching her arms above her head with her eyes closed. Morning light through the round windows, rooftops outside. Full body from the side, sharp focus, natural skin texture.
Candlelit bath, seed 96120
A phone photograph of a nude woman in her twenties with a blonde bob and a slim build, sitting on the edge of a deep stone bathtub surrounded by lit candles, testing the water with one hand and looking down at it. Warm candlelight, dark slate tiles. Full body from the side, sharp focus, natural skin texture.
Oil painting, seed 96240
An oil painting on canvas of a woman in her twenties with auburn hair pinned up loosely, wearing a long white linen dress, standing by a tall open window in a quiet Dutch interior and reading a folded letter, her face turned down toward it. Soft north light falls across her face and the dress, a blue-and-white jug stands on a wooden table and the floor has black and white tiles. Visible brush strokes on her skin, hair and dress, a muted ochre and blue palette. Full body
What it does well
Full-body nudes in daylight, flash and night scenes
Explicit scenes, oral most reliably
Readable text on signs and labels
Anime and painted illustration
Plain-caption prompts: who, what she does, where, the light
Speed: about 12 s for a 1024x1536 image on an 8 GB card with int8 convrot
Image editing with the same files, in ComfyUI's Qwen Image 2.1 edit template
Which file by Vram
8–12 GB: int8 convrot (7.3 GB), the default, almost pixel-identical to bf16
6–8 GB: int4 mixed (4.6 GB), the fastest; compositions shift a little
GGUF Q8_0, Q6_K, Q5_K, Q4_K (7.6 to 4.1 GB), in v1.0
fp8 (7.3 GB): plain fp8 e4m3fn
bf16 and fp16 (14.2 GB): for merging and training; fp16 for GPUs without bf16
Notes
Shibari is VALIDATED, may allow obscure position support don't hesitate to try
中文：基于 Qwen Image 2.1，开箱即可生成裸体与成人场景，无需 LoRA。两个版本：Turbo（6 步，无 CFG）和 Base（完整模型，25–50 步，cfg 3）。8–12G 显存用 int8 convrot，6–8G 用 int4 mixed。仅限非商业用途。
License
Built with Qwen. Qwen is licensed under the Qwen RESEARCH LICENSE AGREEMENT, Copyright (c) 2026 Hangzhou Tongyi Laboratory Technology Co., Ltd. All Rights Reserved. Free, non-commercial use only. Full licence: https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE. LICENSE and NOTICE, with the list of changes, are in the zip.
Description
First release: Qwen Image 2.1 in 4 steps, in nine formats: int8 convrot, int4 mixed, fp8, bf16, fp16 and GGUF Q8_0 / Q6_K / Q5_K / Q4_K. Text encoder and VAE included. 4 steps, cfg 1, euler, simple.
FAQ
Comments (36)
Does it require abliterated/uncesored text encoder - or just the vanilla?
No need for a specific text encoder the vanilla works fine
Those abliterated text encoders do nothing other than destroy images. text encoders aren't censored at all.
this model contains step4 lora that SHIFTS image to left when edit.
any solution for this , the shift is Huge half the pic is gone
For edit i found "sort of" solution. Before passing to "Empty Latent image" node you should multiplify image width dy a number between 1.2-1.6 (use simple math node). You will get wider image but you can crop :-)
use latest v0.2 fixed step5 lora. https://huggingface.co/Viggle/Qwen-Image-2.1-viggle-turbo
The model was updated and the edit shift should be gone
Please release non-turbo version
Upload in progress, look forward to it
Can it also be used for image editing?
Yes the v2 works reliably with editing
They added Qwen 2.1 category, please fix
Thanks for the info, Updated
nice~v2 is better
V3 Base is out now too, curious what you think of it.
有个问题，编辑图片的时候会整体将图片裁切放大。和原始图对不齐。
感谢反馈。这是 V1 Turbo 的问题：编辑时画面会整体偏移、看起来像被裁切放大。从 V2 开始换用了 Viggle v0.2 的蒸馏，这个问题已经修正，V3 也没有。请换用 V2 或 V3（Turbo 或 Base），并使用 ComfyUI 自带的 Qwen Image 2.1 编辑模板。
base int8 version has bug:
结论：这个 safetensors 文件的内部数据偏移表错位了—— 从第 6 个 block 开始，header 记录的文件偏移比实际位置偏了一段，导致 ComfyUI 按错误地址去读 "量化配置" 时，读到的其实是权重数据的中间一段。这是模型转换 / 打包工具的 bug（写文件时中间多写或少写了一块，后续偏移没同步更新）
感谢反馈。我直接检查了 Civitai 上的文件：header 完整（649 个张量，偏移连续，数据正好在文件末尾结束），第 0–8 和第 20 个 block 的量化配置都能正常读取。你下载的文件可能损坏了（下载中断或多线程下载会让某一块之后的数据整体错位）。请核对 SHA256，应为 7C79FAF1B8E279CDF9E60F94C6A3DA70D505CC2AC34DA33377EC5768C03FC69F（Windows：certutil -hashfile NoctQ_V2_base_int8_convrot.safetensors SHA256）。不一致的话请重新下载。int8 文件需要 ComfyUI 0.37.0 或更新版本。
Thanks for the report. I checked the file on Civitai: the header is consistent and the quant config of blocks 0-8 and 20 reads correctly, so your copy is probably damaged. Its SHA256 should be 7C79FAF1B8E279CDF9E60F94C6A3DA70D505CC2AC34DA33377EC5768C03FC69F; if it differs, please download it again. The int8 file needs ComfyUI 0.37.0 or newer.
Very nice, but it needs improvements in specific geometry. I assume those will come soon.
V3 Base is out and it's the one I recommend, worth a try.
Wow, this is crazy... people need to stop bothering with Krea and swap over to Qwen2.1. Can't wait till will get more finetunes coming out. Thanks for this turbo version!
Krea2 is still far better quality in text to image
The Krea2 remains the best in t2i quality. qwen21 good only for editing.
Nope! Krea2 is still undefeated for text to image and remains the king. I’d say anyone insisting on using Qwen2.1 for T2I has probably never tried Krea2 at its best 😂
For editing though qwen21 is the best for local right now.
When i generate 4 people in 4 different pose, Krea 2 generates exactly that first try, Qwen generates a soup of deformed hands and legs while taking 5x more time.
I like Qwen's editing capabilities tho, it might not be a full replacement for Klein 9b for now but it's awesome.
lmao.
please atleast provide example prompt on the image so that people can replicate your results.
Sure, here's one in the style it likes (a plain description: shot type, who, what she does, where, the light): "An amateur photograph of a nude woman in her twenties with curly auburn hair, leaning on the counter of a small old-fashioned kitchen with a cup of coffee, smiling at the camera. Morning sun through the window on the left, plants on the sill. Full body, natural skin texture." Keep the settings of the workflow on the page.
I want to know if this uncensored version works for men?
Very bad quality, noisy image.
Much worse than native or other models.
If that was with a Turbo, try V3 Base, it's the recommended version.
Edit: V2 int4 is also broken.
Edit 2: unlike Turbo v2, int4 v2 works poorly on edit.
Anything i generate with Turbo v2 model is just bad, every single image is noisy and everything in the scene is mushed together (int8), still great for edit tho. quality = base model in edit and with bfs headswap lora it's just great.
Also when i try to download the base model:
You are being rate limited.
The owner of this website has banned you temporarily from accessing this website.
(tried couple of more times and it's downloading again)
The Turbos were experimental, V3 Base is the recommended one now. Glad editing works well for you!