Qwen Image 2512 & 2511 Training Results Are Next Level + 33 ComfyUI Presets for Image & Video Gen

[FurkanGozukara]

Qwen Image 2512 & 2511 Training Results Are Next Level + 33 ComfyUI Presets for Image & Video Gen

Full tutorial: https://www.youtube.com/watch?v=RcoXd9v1t_c

Qwen Image 2512 Text to Image model is a massive upgrade in quality just as Qwen Image Edit 2511 was for image editing tasks based on commands. I have trained both Qwen Image older Base, Qwen Image 2512 new base, Qwen Image 2509 older edit model and Qwen Image 2511 newer edit model and compared them in this video. The results are astonishing. Moreover, we have converted all of our premium SwarmUI image and video generation presets into ComfyUI workflows. Just drag and drop the workflow and start using immediately. All models are downloaded with our premium 1-click to use model downloader app. ComfyUI and SwarmUI are also installed with 1-click installers and our ComfyUI fully supports Sage Attention, Flash Attention, xFormers, Triton and more with RTX 5000 series and all GPUs like 3000, 2000, 4000 series, etc for both Linux and Windows.

📂 Resources & Links:

ComfyUI Installer: [ https://www.patreon.com/posts/ComfyUI-Installers-105023709 ]

SwarmUI Installer, Model Auto Downloader and Presets: [ https://www.patreon.com/posts/SwarmUI-Install-Download-Models-Presets-114517862 ]

SwarmUI & ComfyUI Setup Guide for Windows: [ https://youtu.be/c3gEoAyL2IE ]

SwarmUI & ComfyUI Setup Guide for RunPod & Massed Compute: [ https://youtu.be/bBxgtVD3ek4 ]

Qwen Image Models Training Tutorial: [ https://youtu.be/DPX3eBTuO_Y ]

Wan 2.2 Model Training Tutorial: [ https://youtu.be/ocEkhAsPOs4 ]

Z Image Turbo Model Training Tutorial: [ https://youtu.be/ezD6QO14kRc ]

FLUX Models Fine Tuning / DreamBooth Training Tutorial: [ https://youtu.be/FvpWy1x5etM ]

FLUX Models LoRA Training Tutorial: [ https://youtu.be/nySGu12Y05k ]

Upload / Download Big Files Guide for RunPod & Massed Compute: [ https://youtu.be/X5WVZ0NMaTg ]

⏱️ Video Chapters:

00:00:00 Introduction to Qwen Image Base & 2512 Model Training Findings plus Qwen Image Edit 2509 & 2511 News

00:00:16 Announcement of Training Support for Qwen Image Edit Models & Upcoming Training Guide Overview

00:00:32 New FP8 Quantization Feature in SECourses Musubi Tuner & Full ComfyUI Preset Support Announcement

00:00:48 Converting All SwarmUI Presets into Drag-and-Drop ComfyUI Workflows & HTML Helper File Introduction

00:01:10 How to Use the New HTML File to Identify and Download Necessary Model Bundles via SwarmUI Downloader

00:01:24 Qwen Image Base Model Training Analysis: Observing Overfitting and Quality Degradation at 240 Epochs

00:02:04 Qwen Image 2512 Model Training Results: Massive Quality Boost & Reduced Noise Compared to Base Model

00:02:38 Showcasing New Qwen Image 2512 UHD Realism Presets Available in Both ComfyUI and SwarmUI

00:03:01 Visual Comparison: The Massive Leap in Realism Quality Between Previous Models and New 2512 Model

00:03:38 Instructions on Updating the Training Application via Bat File to Unlock New Qwen Model Support

00:04:03 Selecting Specific Model Versions in the Trainer: Text-to-Image vs Image Edit Plus 2509/2511 Variants

00:04:24 Using the Updated Model Downloader Application to Fetch New Qwen 2512 & 2511 Checkpoints Automatically

00:04:40 Wan 2.2 Training Tutorial Reference & Overview of Recently Updated High-Quality Presets for All Models

00:05:05 Training Dataset Configuration: Using Medium Quality Settings & OHWX Caption Strategy for Best Results

00:05:22 Testing Qwen Image Edit 2509 vs 2511 Models with GTA 5 Style Transfer Training on 120 and 180 Epochs

00:05:46 Analyzing GTA 5 Style Reproduction Accuracy on Both Edit Models Without Using Any Control Images

00:06:11 Style Transfer Experiment: Converting Input Image to Learned Style Using Edit Models Without Image-to-Image

00:06:40 Critical Findings on Edit Models: Lower Epoch Requirements & How Overtraining Alters Input Image Fidelity

00:07:09 Resolution Limitations for Style Transfer: Why 1328x1328 is the Optimal Sweet Spot for Editing Tasks

00:07:30 Reviewing Specific Presets Used for Qwen Image Edit Text-to-Image Generation vs Image Editing Tasks

00:08:02 How to Install and Update ComfyUI with the Latest Standalone Installer Script & Overwriting Old Files

00:08:15 Downloading Model Bundles for ComfyUI Presets Using the SwarmUI Downloader Tool & HTML Guide

00:08:39 Navigating the HTML Guide to Select Correct Model Bundles like Flux 2, Z Image Turbo or Qwen Core

00:09:12 Understanding Preset Limitations: Why Inpainting & Outpainting Workflows Require SwarmUI Editing Tab

00:09:57 Live Demonstration: Dragging and Dropping Flux SRPO Preset into ComfyUI Interface for Instant Setup

00:10:11 Configuring the Model Downloader to Target Your Specific ComfyUI Installation Path for Direct Downloads

00:10:47 Handling Potential VAE Subfolder Issues & Verifying High-Speed Downloads with 100% Hash Verification

00:11:06 Generating Images in ComfyUI: Python 3.10-3.13 Support, RTX 5000/4000/3000 Compatibility & Speed

00:11:54 Recommendation to Use SwarmUI with ComfyUI Backend for Best Experience & Access to All Features

00:12:20 Upcoming Tutorials Teaser: Trellis 2 3D Models & Video/Image Upscaler Guides Coming Soon

Video Transcription

• 00:00:00 Greetings everyone. Today I have got some  important news for you. I have trained with Qwen  

• 00:00:06 Image Base Model and Qwen Image 2512 model and we  have got some very important findings. I also have  

• 00:00:16 trained with Qwen Image Edit 2509 and 2511 models.  I have got some very important news about that as  

• 00:00:25 well. Furthermore, I will show how to train with  newer Qwen Image models. Moreover, we have got a  

• 00:00:32 new FP8 quantization in our SECourses Musubi  Tuner application. This is the very best FP8  

• 00:00:40 quantization. It supports so many new features.  And we now fully support ComfyUI presets. What I  

• 00:00:48 mean by that, we were supporting so many amazing  presets, all the time gets updated with SwarmUI as  

• 00:00:56 you know. Now we support all of them with ComfyUI  as well. I have converted all of the presets into  

• 00:01:04 ComfyUI workflows. So you will be able to just  drag and drop them and use them right away. I also  

• 00:01:10 have prepared a HTML file which shows you how to  download necessary preset models right away with  

• 00:01:18 our SwarmUI downloader. So you will be able to  know which bundle to download for which preset.

• 00:01:24 So let's begin with the Qwen Image new 2512 model  training. I have made fully fresh training both  

• 00:01:33 Qwen Image Base Model and Qwen Image 2512 model.  So the Qwen Image Base Model, this is 120 epoch,  

• 00:01:43 this is 180 epoch, and this is 240 epochs. As  you can see it becomes more overtrained and the  

• 00:01:51 quality drops. This is realism preset, this is  grid images so this is not cherry-pick. You see  

• 00:01:57 degrades overtime. From this to this to this. So  this is the quality of the Qwen Image Base Model,  

• 00:02:04 the previous model. With the Qwen Image 2512  model, we got a huge boost in quality. So this  

• 00:02:12 is 120 epoch. You see it has a lot of noise.  This is 180 epoch. It becomes much better.  

• 00:02:19 This is really good. And this is 240 epoch. It  becomes perfect. You can also choose the middle  

• 00:02:24 ground like 180 epoch but the quality difference,  quality improvement is great compared to the our  

• 00:02:31 older model. So this was our older model realism  preset and this is our newer model. So for newer  

• 00:02:38 model I have used our newer preset Qwen Image  2512 UHD Realism. Also it shows the date as  

• 00:02:45 well. Now this also exist in our ComfyUI preset.  You can see that Qwen Image 2512 UHD High Realism  

• 00:02:53 and Qwen Image 2512 UHD Realism preset. So you  can use this preset in ComfyUI or in SwarmUI.

• 00:03:01 And we got a huge boost in realism. So this is  120 epoch, this is 180 epoch, and this is 240  

• 00:03:08 epoch. You see the realism quality, it is just top  level. And these are just grid images. 120 epoch,  

• 00:03:15 180 epoch, 240 epoch. You see the quality jump  from previous model to new model is just next  

• 00:03:24 level. This was our previous model high realism  and this is our new model high realism. It is  

• 00:03:31 just amazingly better compared to the previous  model. So how you can train with newer model? Just  

• 00:03:38 download the latest training zip file. The link  will be in the description of the video. Overwrite  

• 00:03:43 previous files, run the windows install and update  .bat file. Then when you run the application in  

• 00:03:50 the Qwen, you will see that now we support this  way model version. So you can pick the Qwen  

• 00:03:57 Image older and newer model. This is text to image  model, this model, text to image model. Or you can  

• 00:04:03 pick the Qwen Image Edit Plus 2509 and Qwen Image  Edit 2511. 2511 is better. If you don't know how  

• 00:04:11 to train, we have a full tutorial. It is here. The  link will be also in the description of the video.  

• 00:04:16 So exactly same, nothing changed. You just follow  this video and now our model downloader supports  

• 00:04:24 new models 2512 and 2509. You see when you start  the model downloader you will see the newer 2512  

• 00:04:33 model and 2511 model like this. We also now have  Wan 2.2 training tutorial too. If you want you  

• 00:04:40 can also watch this Wan 2.2 training tutorial and  follow it. I have updated all the presets recently  

• 00:04:47 so these are all up to date with highest quality.  You can also read their descriptions and all of  

• 00:04:53 these presets are also now available with ComfyUI.  You know our ComfyUI installer is standalone  

• 00:04:59 installer. It is not mandatory to use it with  SwarmUI. Also for training I used our classical  

• 00:05:05 dataset. This is medium quality and the training  caption was just OHWX as in the tutorial video.

• 00:05:12 For testing the Qwen Image Edit 2509 and 2511, I  used our GTA 5 style. So these first two images  

• 00:05:22 are for the base models. You see this is Qwen  2509, this is Qwen 2511. So this is what the  

• 00:05:29 model generates without training. And as we train,  you see this is 09, this is 11, 120 epochs. This  

• 00:05:37 is 09, this is 11, 180 epoch and you can see  that both model generates GTA 5 style exactly.  

• 00:05:46 Perfect. Perfect quality. And this is 2509 and  this is 2511 Qwen Image Edit model. I didn't  

• 00:05:55 use control images so these are just trained with  the training images, nothing else. So both model  

• 00:06:02 perfectly learns. In my opinion as expected, the  2511 model is better. Also it has an advantage.  

• 00:06:11 What advantage it has? So I made a test about  converting my image into learned style image.  

• 00:06:19 How did I do that? I gave this image as an input  image, as a prompt image and I said that convert  

• 00:06:25 it into OHWX style, nothing else. So this is not  image to image. This is Qwen Image Edit model  

• 00:06:32 capability converting image. I did a big test and  what I noticed is that as you train more epochs,  

• 00:06:40 it starts to change input image itself. So this  is more loyal to the input image and as I train  

• 00:06:47 more becomes more like this, more changed, more  overtrained. So the Qwen Image Edit model not  

• 00:06:54 requires too many epochs as the Qwen Image Base  text to image model. This is what I have found.  

• 00:07:01 Moreover, as you go with higher resolution, this  model loses its stylization editing capability. So  

• 00:07:09 this is very high resolution and as you can see it  is not that good. So the best resolution is 1328  

• 00:07:15 to 1328 for this task. But when we generate images  with our preset we are upscaling and it is working  

• 00:07:24 perfect but for image editing task it didn't  work that well for style transfer like this.

• 00:07:30 So for Qwen Image Edit models I have used the  following presets. So this is text to image  

• 00:07:38 generation with Qwen Image Edit model with  stylization or with text to image model with  

• 00:07:43 stylization. And for image editing I used this  preset as I have shown in the previous tutorials.  

• 00:07:50 Again just update the application and you can use  these options to select which model you are going  

• 00:07:55 to train. So what about how to use ComfyUI?  So for using ComfyUI, download the latest zip  

• 00:08:02 file. The link will be in the description of the  video. Overwrite older files. Just run the windows  

• 00:08:07 install or update comfyui.bat file. This will  install and update your file. To download models  

• 00:08:15 as a preset, as a bundle, you need to have SwarmUI  model downloader application. The link will be in  

• 00:08:21 the description of the video. Download the latest  one. And start the windows start download models  

• 00:08:26 app.bat file. In the latest ComfyUI zip file  you will see the presets and the presets are  

• 00:08:31 like this. So in this presets file open the which  bundles download preset. And according to the  

• 00:08:39 model preset that you want to use, let's say you  want to use FLUX 2, download the necessary bundle.  

• 00:08:45 We have bundles for major models. You see we have  Z Image Turbo Core bundle, FLUX 2 Low Ram bundle,  

• 00:08:53 FLUX 2 Core bundle, Qwen Image Core bundle. When  you open them you will see all the models it  

• 00:08:59 downloads. When you download these models you  will have all the models downloaded for using  

• 00:09:04 the preset. Or Wan 2.2 Core bundle. Moreover you  can see which bundles covering which presets from  

• 00:09:12 here. We support all the presets in SwarmUI except  2 presets which are outpainting and inpainting.  

• 00:09:20 Because these two presets depending on the image  editing of the SwarmUI. And we also have a full  

• 00:09:27 tutorial for that. When you watch this tutorial,  Qwen Image Models Realism, it shows you how to do  

• 00:09:33 inpainting and outpainting. So I recommend  to follow this video. For Qwen Image Edit  

• 00:09:38 newer model follow this video. We have videos for  everything. You can just leave a comment, message  

• 00:09:44 me from Patreon or from Discord and I can show you  which tutorial you should follow for which task.

• 00:09:50 So let's make a demonstration of the ComfyUI.  So I will drag and drop my FLUX SRPO. And it  

• 00:09:57 has selected the FLUX SRPO model. Currently my  ComfyUI see all of my models inside SwarmUI and  

• 00:10:05 ComfyUI folder. By the way when you download for  ComfyUI, this is how you download the models. Go  

• 00:10:11 to your ComfyUI installation, go to ComfyUI, go to  models folder, copy the path like this. Paste the  

• 00:10:18 path here. So you see you paste this models  path, select this ComfyUI folder structure,  

• 00:10:24 then the rest is same. Just click download and  it will download all the models. For example  

• 00:10:28 let's download Z Image Turbo Core bundle. It will  start downloading right away with the fastest way,  

• 00:10:34 with the hash verification. So all of my downloads  are 100 percentage accurate. You see the speed is  

• 00:10:41 100 megabytes per second. I will just cancel  it since I have them. And you can also choose  

• 00:10:47 different models from here. You may have issues  with the VAE because the VAE are downloaded into  

• 00:10:54 subfolders in SwarmUI. So just pick the VAE  from here if you don't see it immediately.  

• 00:11:00 But it will see the others. So that's it. You  just type your prompt, you change your output  

• 00:11:06 resolution. Currently it is set 1024. I just hit  generate. Everything will work right away. I have  

• 00:11:13 tested all the presets that I have here. So all of  them is working 100 percentage with our installers  

• 00:11:20 because some of the presets depending on some  of the custom extensions. Therefore you have to  

• 00:11:26 use our installer for ComfyUI and our ComfyUI  installer supports Python 3.10, Python 3.11,  

• 00:11:33 Python 3.12, and Python 3.13 with Sage Attention,  Flash Attention, Triton, all the necessary  

• 00:11:41 libraries. It works perfect on RTX 5000 series  GPUs or all of the GPUs like 4000, 3000, 2000.  

• 00:11:48 I support all of them. And the image has been  generated like this. So this is how you use the  

• 00:11:54 ComfyUI presets. They are all set. You don't need  to do anything. You can download all the models  

• 00:12:01 with our downloader. You can find which models  are needed here. And the best part is you can use  

• 00:12:08 SwarmUI with our ComfyUI installation as a backend  and use everything in SwarmUI which I recommend.  

• 00:12:14 Because SwarmUI is the best UI right now, most up  to date UI right now to do everything. And we have  

• 00:12:20 tutorials for everything like image editing or  inpainting, outpainting, fixing errors, whatever  

• 00:12:28 you can imagine for. And we have tutorials for  all the trainings. Hopefully Trellis 2 tutorial  

• 00:12:35 is coming and the video and image upscaler  tutorial is coming. So these are the news today  

• 00:12:41 I wanted to show you. Stay subscribed, ask me  any questions you want. Hopefully see you later.