MCPcopy Index your code
hub / github.com/ModelTC/LightX2V-Qwen-Image-Lightning

github.com/ModelTC/LightX2V-Qwen-Image-Lightning @main

Chat with this repo
repository ↗ · DeepWiki ↗ · + Follow
1 symbols 14 edges 1 files 0 documented · 0%
What it actually does AI analysis from the code graph — generated when you open this
loading…
README

Qwen-Image-Lightning

We are excited to release the distilled version of Qwen-Image. It preserves the capability of complex text rendering.

🔥 Latest News

📑 Community Support

📑 Todo List

  • [x] Qwen-Image-Lightning-4/8steps-V1.x
  • [x] ComfyUI Workflow
  • [x] Qwen-Image-Edit-Lightning-4/8steps-V1.0
  • [x] Qwen Edit ComfyUI Workflow
  • [x] Qwen-Image-Lightning-4/8steps-V2.0
  • [ ] Qwen-Image-Edit-Lightning-4/8steps-V2.0
  • [x] Qwen-Image-Edit-2509-Lightning-4/8steps-V1.0
  • [x] Qwen Edit 2509 ComfyUI Workflow
  • [x] Qwen-Image-Edit-2511-Lightning-4steps-V1.0
  • [ ] Qwen Edit 2511 ComfyUI Workflow
  • [x] Qwen-Image-2512-Lightning-4steps-V1.0

📑 Comparison between V1.x and V2.x

Compared to V1.0, V2.0 produces images with reduced over-saturation, resulting in improved skin texture and more natural-looking visuals.

Qwen-Image-Lightning

4steps-V1.0 NFE=4 4steps-V2.0 NFE=4 8steps-V1.1 NFE=8 8steps-V2.0 NFE=8
111 112 113 114
111 112 113 114

📑 Using Lightning LoRAs with FP8 Models

The utilization of existing LoRA weights with the qwen_image_fp8_e4m3fn.safetensors base model results in the grid artifacts reported in Issue #32, wherein generated images exhibit a grid-like pattern. This artifact stems from the fact that the qwen_image_fp8_e4m3fn.safetensors model was produced by directly downcasting the original bf16 weights, rather than employing a calibrated conversion process with appropriate scaling.

In response to the prevalent use of this model, we have released a new set of Lightning LoRA weights specifically optimized for this base. These weights were generated by distilling the qwen_image_fp8_e4m3fn.safetensors model using bf16 guidance, thereby mitigating the artifact issue.

An alternative solution we provide is a newly released FP8 base weight, converted with scaling to ensure better compatibility with the existing suite of LoRA weights trained on BF16 bases. A comparative analysis is provided below.

Base + LoRA bf16 base + LoRA trained on bf16 base fp8 base + LoRA trained on bf16 base fp8 base + LoRA trained on fp8 base scaled fp8 base + LoRA trained on bf16 base
Correctness ✅ ❌ ✅ ✅
Comparison 111 112 113 114
Workflow base-bf16-lora-on-bf16.json base-fp8-lora-on-bf16.json base-fp8-lora-on-fp8.json base-scaled-fp8-lora-on-bf16.json

📑 T2I Performance Report

To assess the distilled models' performance characteristics, including their strengths and limitations, we compare the performance of the three models, i.e., Qwen-Image, Qwen-Image-Lightning-8steps-V1.1, and Qwen-Image-Lightning-4steps-V1.0, in different scenarios. The results can be reproduced following the section below.

- Quality and Speed

Compared to the base model, the distilled models (8-step and 4-step) deliver a 12–25× speed improvement with no significant loss in performance in most cases.

Prompt Base NFE=100 8steps-V1.1 NFE=8 4steps-V1.0 NFE=4
一个会议室，墙上写着"3.14159265-358979-32384626-4338327950"，一个小陀螺在桌上转动。 111 112 113
宫崎骏的动漫风格。平视角拍摄，阳光下的古街热闹非凡。一个穿着青衫、手里拿着写着“阿里云”卡片的逍遥派弟子站在中间。旁边两个小孩惊讶的看着他。左边有一家店铺挂着“云存储”的牌子，里面摆放着发光的服务器机箱，门口两个侍卫守护者。右边有两家店铺，其中一家挂着“云计算”的牌子，一个穿着旗袍的美丽女子正看着里面闪闪发光的电脑屏幕；另一家店铺挂着“云模型”的牌子，门口放着一个大酒缸，上面写着“千问”，一位老板娘正在往里面倒发光的代码溶液。 121 122 123
一副典雅庄重的对联悬挂于厅堂之中，房间是个安静古典的中式布置，桌子上放着一些青花瓷，对联上左书“义本生知人机同道善思新”，右书“通云赋智乾坤启数高志远”， 横批“智启通义”，字体飘逸，中间挂在一着一副中国风的画作，内容是岳阳楼。 131 132 133
A movie poster. The first row is the movie title, which reads “Imagination Unleashed”. The second row is the movie subtitle, which reads “Enter a world beyond your imagination”. The third row reads “Cast: Qwen-Image”. The fourth row reads “Director: The Collective Imagination of Humanity”. The central visual features a sleek, futuristic computer from which radiant colors, whimsical creatures, and dynamic, swirling patterns explosively emerge, filling the composition with energy, motion, and surreal creativity. The background transitions from dark, cosmic tones into a luminous, dreamlike expanse, evoking a digital fantasy realm. At the bottom edge, the text “Launching in the Cloud, August 2025” appears in bold, modern sans-serif font with a glowing, slightly transparent effect, evoking a high-tech, cinematic aesthetic. The overall style blends sci-fi surrealism with graphic design flair—sharp contrasts, vivid color grading, and layered visual depth—reminiscent of visionary concept art and digital matte painting, 32K resolution, ultra-detailed. 141 142 143
一张企业级高质量PPT页面图像，整体采用科技感十足的星空蓝为主色调，背景融合流动的发光科技线条与微光粒子特效，营造出专业、现代且富有信任感的品牌氛围；页面顶部左侧清晰展示橘红色Alibaba标志，色彩鲜明、辨识度高。主标题位于画面中央偏上位置，使用大号加粗白色或浅蓝色字体写着“通义千问视觉基础模型”，字体现代简洁，突出技术感；主标题下方紧接一行楷体中文文字：“原生中文·复杂场景·自动布局”，字体柔和优雅，形成科技与人文的融合。下方居中排布展示了四张与图片，分别是：一幅写实与水墨风格结合的梅花特写，枝干苍劲、花瓣清雅，背景融入淡墨晕染与飘雪效果，体现坚韧不拔的精神气质；上方写着黑色的楷体"梅傲"。一株生长于山涧石缝中的兰花，叶片修长、花朵素净，搭配晨雾缭绕的自然环境，展现清逸脱俗的文人风骨；上方写着黑色的楷体"兰幽"。一组迎风而立的翠竹，竹叶随风摇曳，光影交错，背景为青灰

Core symbols most depended-on inside this repo

main
called by 1
generate_with_diffusers.py

Shape

Function 1

Languages

Python100%

Modules by API surface

generate_with_diffusers.py1 symbols

For agents

$ claude mcp add LightX2V-Qwen-Image-Lightning \
  -- python -m otcore.mcp_server <graph>

⬇ download graph artifact