<div data-youtube-video="https://youtu.be/4YEU8Ui6u1w"><iframe src="https://www.youtube.com/embed/4YEU8Ui6u1w" width="853" height="480" allowfullscreen></iframe></div> 
 
Qwen Image 2.1 Fix 1.0 / 2.0 image quality enhancement workflow. Removes the GPT-generated look and high-frequency fake details while keeping the original composition, delivering cleaner, more realistic, more textured results. 
 
Main features: 
- Built on Qwen Image 2.1 base (Qwen-Image_2.1 + qwen3vl_8b CLIP + qwen_image_2.1 VAE) with the detail-fix LoRA (qwen2.1-detail-fix-opinionated v1.0) 
- Switch-controlled dual output: toggle between Fix 1.0 and Fix 2.0 strength in the same graph 
- Each pass: 20 steps, CFG 3.5, euler/simple, randomized seed 
- rgthree Image Comparer panels put the original next to Fix 1.0 and Fix 2.0 for direct judging on the same subject (roses, sculptures, water scenes, portraits in the tutorial) 
- Prompt construction helpers (TextGenerate / RegexExtract / StringConcatenate) keep the cleaned image on-prompt 
 
Suggested workflow: 
1. Load the image to be cleaned up and set the prompt 
2. Run the graph; original, Fix 1.0 and Fix 2.0 render for side-by-side comparison 
3. Toggle the switch to pick the fix level per image type (portraits vs landscapes vs objects) and save 
 
Dependencies: 
- Qwen-Image_2.1.safetensors (base UNET) 
- qwen3vl_8b_int8_convrot.safetensors (CLIP) 
- qwen_image_2.1_vae_bf16.safetensors (VAE) 
- qwen2.1-detail-fix-opinionated-v1.0.safetensors (detail-fix LoRA) 
- rgthree Node Suite (Image Comparer), ComfySwitchNode 
 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2109547621848211457?inviteCode=rh-v1111 
Bilibili tutorial video: https://www.bilibili.com/video/BV14zp96REo9/ 
RunningHub registration with invite code rh-v1111 (1000 RH coins): https://www.runninghub.ai/?inviteCode=rh-v1111

Qwen Image 2.1 Fix 1.0/2.0 Quality Repair

<div data-youtube-video="https://youtu.be/4YEU8Ui6u1w"></div> 
 
Qwen Image 2.1 Fix 1.0 / 2.0 image quality enhancement workflow. Removes the GPT-generated look and high-frequency fake details while keeping the original composition, delivering cleaner, more realistic, more textured results. 
 
Main features: 
- Built on Qwen Image 2.1 base (Qwen-Image_2.1 + qwen3vl_8b CLIP + qwen_image_2.1 VAE) with the detail-fix LoRA (qwen2.1-detail-fix-opinionated v1.0) 
- Switch-controlled dual output: toggle between Fix 1.0 and Fix 2.0 strength in the same graph 
- Each pass: 20 steps, CFG 3.5, euler/simple, randomized seed 
- rgthree Image Comparer panels put the original next to Fix 1.0 and Fix 2.0 for direct judging on the same subject (roses, sculptures, water scenes, portraits in the tutorial) 
- Prompt construction helpers (TextGenerate / RegexExtract / StringConcatenate) keep the cleaned image on-prompt 
 
Suggested workflow: 
1. Load the image to be cleaned up and set the prompt 
2. Run the graph; original, Fix 1.0 and Fix 2.0 render for side-by-side comparison 
3. Toggle the switch to pick the fix level per image type (portraits vs landscapes vs objects) and save 
 
Dependencies: 
- Qwen-Image_2.1.safetensors (base UNET) 
- qwen3vl_8b_int8_convrot.safetensors (CLIP) 
- qwen_image_2.1_vae_bf16.safetensors (VAE) 
- qwen2.1-detail-fix-opinionated-v1.0.safetensors (detail-fix LoRA) 
- rgthree Node Suite (Image Comparer), ComfySwitchNode 
 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2109547621848211457?inviteCode=rh-v1111 
Bilibili tutorial video: https://www.bilibili.com/video/BV14zp96REo9/ 
RunningHub registration with invite code rh-v1111 (1000 RH coins): https://www.runninghub.ai/?inviteCode=rh-v1111

Files with SHA256: f13ae72683fe56d6fbe4e09544e955ca28c17af151f85995dffd92abab05b8ee