<h1 id="qwen3-vl">Qwen3-VL</h1>Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.This generation delivers comprehensive upgrades across the board: superior text understanding &amp; generation, deeper visual perception &amp; reasoning, extended context length, enhanced spatial and video dynamics comprehension, and stronger agent interaction capabilities.Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment.Key Enhancements:<ul><li>Visual Agent: Operates PC/mobile GUIs—recognizes elements, understands functions, invokes tools, completes tasks.</li><li>Visual Coding Boost: Generates Draw.io/HTML/CSS/JS from images/videos.</li><li>Advanced Spatial Perception: Judges object positions, viewpoints, and occlusions; provides stronger 2D grounding and enables 3D grounding for spatial reasoning and embodied AI.</li><li>Long Context &amp; Video Understanding: Native 256K context, expandable to 1M; handles books and hours-long video with full recall and second-level indexing.</li><li>Enhanced Multimodal Reasoning: Excels in STEM/Math—causal analysis and logical, evidence-based answers.</li><li>Upgraded Visual Recognition: Broader, higher-quality pretraining is able to “recognize everything”—celebrities, anime, products, landmarks, flora/fauna, etc.</li><li>Expanded OCR: Supports 32 languages (up from 19); robust in low light, blur, and tilt; better with rare/ancient characters and jargon; improved long-document structure parsing.</li><li>Text Understanding on par with pure LLMs: Seamless text–vision fusion for lossless, unified comprehension.</li></ul>Model Architecture Updates:<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_arc.jpg" width="80%" /><ol><li>Interleaved-MRoPE: Full‑frequency allocation over time, width, and height via robust positional embeddings, enhancing long‑horizon video reasoning.</li><li>DeepStack: Fuses multi‑level ViT features to capture fine‑grained details and sharpen image–text alignment.</li><li>Text–Timestamp Alignment: Moves beyond T‑RoPE to precise, timestamp‑grounded event localization for stronger video temporal modeling.</li></ol>

Qwen3-VL

qwen3vl_8b_bf16.safetensors

qwen3vl_8b_int8_convrot.safetensors

qwen3vl_8b_w4a8.safetensors

qwen3vl_8b_nvfp4.safetensors

qwen3vl_8b_fp8_scaled.safetensors

Repost From Huggingface <a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1">https://huggingface.co/Qwen/Qwen-Image-2.1</a><h2 id="introduction">Introduction</h2>We are excited to open-source Qwen-Image-2.1, a unified text-to-image generation and image editing model in the Qwen family. With just 7B parameters in its visual generation component (32 Single-Stream DiT layers), Qwen-Image-2.1 balances generation quality, inference efficiency, and versatility.Four key improvements define this release:<ul><li>Compact and Efficient — A lightweight architecture with mixed-granularity attention and prefix KV cache reuse delivers strong image quality at low computational cost.</li><li>Native Transparency, Unified Creation and Editing — Generate regular or transparent (RGBA) images from text, edit transparent layers, and extract subjects from photographs—all in one model.</li><li>Versatile Editing — Support up to 10 reference images, specify local edits via circles, painted annotations, or separate masks, and preserve identity for people and products.</li><li>Realistic Textures and Refined Aesthetics — Improved typography, portrait lighting, and fine details for more visually compelling results.</li></ul>

Qwen Image 2.1

qwenImage21_21_txt_bf16.safetensors

qwenImage21_21_txt_int8.safetensors

qwenImage21_21_bf16.safetensors

qwenImage21_21_int8.safetensors

qwen_image_2.1_vae_bf16.safetensors

<h3 id="about-these-variants">About these variants</h3>Four precision levels are available for Qwen-Image-2.1, trading VRAM and speed against generation quality:- BF16 — full-precision reference (~14 GB). Highest quality, largest footprint. - INT8 (W8A8) — 8-bit weights + 8-bit activations (~7 GB). Near-lossless quality, ~2× smaller than BF16, runs on INT8 tensor cores (RTX 30-series and up). - INT6 (W6A8) — 6-bit weights + 8-bit activations (~5.5 GB). Middle-ground: smaller than INT8 with only a small quality trade-off. - INT4 (W4A8) — 4-bit weights + 8-bit activations (~4 GB). Smallest footprint; uses ConvRot with a per-tensor codebook that decodes to INT8 for compute, so it runs on the same INT8 hardware as W8A8. Larger quality trade-off than INT6. Qwen-Image-2.1 7B — INT4 (W4A8) &amp; INT6 (W6A8) ConvRot for ComfyUI INT4 and INT6 quantized weights of Qwen-Image-2.1 for fast, low-VRAM inference in ComfyUI. This is a modified (quantized) version of the Qwen-Image-2.1 model. It is not an official Qwen release and is not endorsed by the Qwen team. About Qwen-Image-2.1 A unified text-to-image generation and image editing model in the Qwen family. With just 7B parameters in its visual generation component (32 Single-Stream DiT layers), Qwen-Image-2.1 balances generation quality, inference efficiency, and versatility. Highlights - Efficient Image Generation — combines strong visual performance with fast inference and a compact design, making high-quality image creation accessible across a wide range of creative workflows. - Flexible Creative Control — supports diverse inputs, outputs, and localized edits, giving creators the flexibility to explore ideas and refine details within a unified workflow. Key improvements in 2.1 - Compact and Efficient — lightweight architecture with mixed-granularity attention and prefix KV cache reuse delivers strong image quality at low computational cost. - Native Transparency, Unified Creation and Editing — generate regular or transparent (RGBA) images from text, edit transparent layers, and extract subjects from photographs — all in one model. - Versatile Editing — support up to 10 reference images, specify local edits via circles, painted annotations, or separate masks, and preserve identity for people and products. - Realistic Textures and Refined Aesthetics — improved typography, portrait lighting, and fine details for more visually compelling results. License Qwen-Image-2.1 is licensed under the [<a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE">Qwen Research License</a>]<a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE"> </a>These files are a quantized derivative and are distributed under the same license — see the original repository for full terms, permitted uses, and any commercial-use restrictions.

Qwen Image 2.1/Turbo INT4(W4A8)/INT6(W6A8)/INT8

qwenImage21INT8INT4_int8_txt.safetensors

qwenImage21INT8INT4_int8.safetensors

INT8

Turbo INT4(W4A8)

Turbo INT6(W6A8)

Turbo INT8

Base INT6(W6A8)

Base INT4 (W4A8)

<h2 id="noct-q">Noct Q</h2>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.<h3 id="what's-new-in-v3">What's new in V3</h3><ul><li>Turbo rebuilt and sampled in 6 steps: sharper images and seeds as varied as the full model</li><li>In a blind test on the same 150 prompts and seeds as V2 Turbo, fewer defects with the same prompt adherence and explicit rendering</li><li>Base is a new model, stronger on explicit scenes</li></ul><h3 id="settings-v3">Settings V3</h3><ul><li>Turbo: 6 steps, euler, no CFG, 832x1248.</li><li>Base: 25–50 steps, euler, simple. V3 Base: cfg 3 with the negative prompt, both set in the workflow.</li></ul><h3 id="prompt-examples">Prompt examples</h3>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<blockquote>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.</blockquote>Candlelit bath, seed 96120<blockquote>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.</blockquote>Oil painting, seed 96240<blockquote>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</blockquote><h3 id="what-it-does-well">What it does well</h3><ul><li>Full-body nudes in daylight, flash and night scenes</li><li>Explicit scenes, oral most reliably</li><li>Readable text on signs and labels</li><li>Anime and painted illustration</li><li>Plain-caption prompts: who, what she does, where, the light</li><li>Speed: about 12 s for a 1024x1536 image on an 8 GB card with int8 convrot</li><li>Image editing with the same files, in ComfyUI's Qwen Image 2.1 edit template</li></ul><h3 id="which-file-by-vram">Which file by Vram</h3><ul><li>8–12 GB: int8 convrot (7.3 GB), the default, almost pixel-identical to bf16</li><li>6–8 GB: int4 mixed (4.6 GB), the fastest; compositions shift a little</li><li>GGUF Q8_0, Q6_K, Q5_K, Q4_K (7.6 to 4.1 GB), in v1.0</li><li>fp8 (7.3 GB): plain fp8 e4m3fn</li><li>bf16 and fp16 (14.2 GB): for merging and training; fp16 for GPUs without bf16</li></ul><h3 id="notes">Notes</h3><ul><li>Shibari is VALIDATED, may allow obscure position support don't hesitate to try</li></ul>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。<h3 id="license">License</h3>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: <a target="_blank" rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE">https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE</a>. LICENSE and NOTICE, with the list of changes, are in the zip.

Noct Q - Uncensored Realism (built on Qwen Image 2.1) | 夜曲 ✨

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.

NoctQ_V1_Q6_K.gguf

NoctQ_V1_bf16.safetensors

noctQUncensoredRealismBuiltOn_v10_txt.safetensors

NoctQ_V1_fp8.safetensors

NoctQ_V1_Q8_0.gguf

NoctQ_V1_fp16.safetensors

NoctQ_V1_licence_notice_workflow.zip

NoctQ_V1_int8_convrot.safetensors

NoctQ_V1_int4_mixed.safetensors

NoctQ_V1_Q5_K.gguf

NoctQ_V1_Q4_K.gguf

v1.0

V4.0 Base

V3.0 Base

V3.0 Turbo (Experimental)

V2.0 Turbo (Experimental)

V2.0 Base

<h2 id="qwen-image-2.1-text-encoder">Qwen Image 2.1 Text Encoder</h2>The original text encoder of Qwen Image 2.1 (Qwen3-VL 8B), ready for ComfyUI.<h3 id="what-it-is">What it is</h3><ul><li>The model that reads your prompt for Qwen Image 2.1: Qwen3-VL 8B</li><li>Two files<ul><li>int8 convrot, 9.35 GB, the default </li><li>w4a8, 6.3 GB, for 8 GB cards and low RAM </li></ul></li><li>Uncensored out of the box</li><li>Same file names as in ComfyUI's official templates, so they load as is</li></ul>No abliteration neededQwen Image 2.1 renders nudity and adult scenes with this original encoder. Noct Q uses it.<h3 id="works-with">Works with</h3><ul><li>Qwen Image 2.1, base and turbo, for text-to-image and editing</li><li>Noct Q, the uncensored Qwen Image 2.1 in Turbo and Base versions: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2958896">https://civarchive.com/models/2958896</a></li><li>The Qwen Image 2.1 VAE: <a target="_blank" rel="ugc" href="https://civarchive.com/models/2960925">https://civarchive.com/models/2960925</a></li><li>Not with Krea 2, which uses the 4B encoder: <a rel="ugc" href="https://civarchive.com/models/2974755">Krea 2 Text Encoder</a></li></ul><h3 id="install">Install</h3><ul><li>Put it in ComfyUI/models/text_encoders/ and load it with the Load CLIP node, type qwen_image (ComfyUI 0.37.0 or newer)</li></ul><h3 id="about-the-images">About the images</h3>Rendered with Qwen Image 2.1 Base (int8 convrot, 25 steps, cfg 1, 1024x1536) and this encoder. Each prompt tests one thing the encoder has to get right: text in English and Chinese, counts, placement, two people with their own clothes.中文：Qwen Image 2.1 原版文本编码器（Qwen3-VL 8B）：int8 convrot（9.35 GB，默认）和 w4a8（6.3 GB，适合 8G 显存）。放入 ComfyUI/models/text_encoders/，用 Load CLIP 节点（type 选 qwen_image）加载。Qwen Image 2.1 使用原版编码器即可生成成人内容，无需 abliterated 版本。仅限非商业用途。<h3 id="license">License</h3>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. LICENSE and NOTICE are in the zip.

Qwen Image 2.1 Text Encoder (original Qwen3-VL 8B, uncensored) | 夜曲 ✨

The original Qwen3-VL 8B text encoder of Qwen Image 2.1, in int8 convrot and w4a8. Built with Qwen.

qwen_image_2.1_text_encoder_licence_notice.zip

Originally released by <a rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1">Qwen on Hugging Face</a>. All credit goes to the Qwen team.
Qwen Image 2.1 combines text-to-image generation and image editing in a single model. Its visual generation component has 7 billion parameters across 32 single-stream DiT layers, balancing image quality, efficiency and versatility.
The release introduces four main improvements:
<ul>
 <li>Efficient architecture: Mixed-granularity attention and reuse of cached prefixes reduce computation while maintaining image quality.</li>
 <li>Generation and editing with transparency: Create regular or transparent RGBA images, edit transparent layers and isolate subjects from photographs.</li>
 <li>Flexible editing: Work with up to ten reference images, guide local changes using circles, painted annotations or separate masks, and preserve the identity of people and products.</li>
 <li>Improved visual detail: Refined typography, portrait lighting, realistic textures and finer details.</li>
</ul>
Released under the <a rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1/blob/main/LICENSE">Qwen Research License Agreement</a>. See the <a rel="ugc" href="https://huggingface.co/Qwen/Qwen-Image-2.1">original model card</a> for examples and usage instructions.

Files with SHA256: 8bfd0f6e12abf2d2d697ecc888e5e90b0d6741d6708f05799f53afa560452e8f