Home Creators Posts Import Register Favorites Logout

Content

Qwen Image compared with FLUX Dev and FLUX Krea over 700 generations; Qwen follows prompts far better and has a 68% bigger native resolution than FLUX Dev.

Compares the Qwen Image Realism and Qwen Image Max Quality settings; the presets (High Quality and Realism Fast) and full grids are in the SwarmUI post. The model runs on GPUs with as low as 6 GB VRAM when you have enough RAM, and even Q4_1 gives good results.

  • I have done over 700 generations to find out the very best configuration for generating the very best images in Qwen Image model. I have shared the presets and the grid images in below post

  • https://www.patreon.com/posts/114517862

    • Our presets are made after extremely carefuly analysis of all experiments

    • This above link zip file also has downloader to auto download all Qwen Image models and more

    • Used 8x A6000 GPU having machine to conduct the experiments but Qwen Image model works as low as 6 GB VRAM GPUs if you have RAM

  • When you analyze the grids you would also see that even Q4_1 of Qwen Image yields amazing results. If you have RAM even 6 GB GPUs can use this model to generate amazing images.

=======

Patreon exclusive posts index to find our scripts easily, Patreon scripts updates history to see which updates arrived to which scripts and amazing Patreon special generative scripts list that you can use in any of your task.

Join discord to get help, chat, discuss and also tell me your discord username to get your special rank : SECourses Discord

Please also Star, Watch and Fork our Stable Diffusion & Generative AI  GitHub repository and join our Reddit subreddit and follow me on LinkedIn (my real profile)

=======

  • How to install ComfyUI with Flash Attention, Sage Attention, xFormers, Triton, DeepSpeed, Insightface, onnxruntime-gpu, GGUF tutorial shared in below video

  • How to install SwarmUI and use existing ComfyUI as a backend and use our presets tutorial share in below video

------

How to Use ComfyUI, SwarmUI and Qwen Image and More

  • It is mandatory to update your ComfyUI and SwarmUI before using Qwen Image model

    • To update use latest zip files and run Windows_Update_SwarmUI and Windows_Update_ComfyUI bat files

  • If you want to use ComfyUI or SwarmUI on RunPod below is a fully up to date tutorial

  • If you want to use ComfyUI or SwarmUI on Massed Compute, which I recommend, below is a fully up to date tutorial

    • https://youtu.be/8cMIwS9qo4M

      • Check video chapters to find relevant sections if you want like : 36:08 Massed Compute - Advanced: Integrating with the Pre-installed SwarmUI

-----------

Qwen Image Model Analysis

  • Qwen Image prompt understanding and following is extremely superior compared to FLUX Dev

  • It realism is also better than FLUX Dev but not at the level of FLUX Krea Dev level

    • However, with LoRAs or Fine Tuning there is 0 doubt that it will surpass the FLUX model

  • Moreover, default resolution of the Qwen Image is 1328x1328 (1,763,584) pixel compared to 1024x1024 (1,048,576) pixel of FLUX Dev

    • So Qwen Image has 68% bigger resolution than FLUX Dev which is a very serious improvement in native resolution and quality

  • We have 2 presets for for Qwen Image

    • Qwen Image High Quality : Follows prompt better and generates complex scenes better (takes exactly 2x time of Fast preset)

    • Qwen Image Realism Fast : Generates more realistic output

    • Moreover our presets are robust and therefore they work between 20-50 steps fairly well

    • As you do more steps better but 20 steps can be also acceptable level

  • Hopefully Qwen Image training coming to Kohya Musubi Tuner and I will make a 1-click to install and run Musubi Tuner GUI for Qwen Image training and research and find out the very best traning parameters and prepare configs for every GPU

    • I expect that you will be able to train Qwen Image as low as 6 GB VRAM if you have RAM

  • Check out the above post images clearly to see the difference between Qwen vs FLUX - they are highest quality

Tutorials Arsenal

Our Installers

  • Our installers always automatically generates an isolated Python VENV and install libraries into that

  • Thus, they will never break your computer or existing running applications

Related posts

All posts in this topic: Research, Comparisons and Experiments · New to SECourses? Start Here: Best and Most Popular SECourses Posts

Files

Comments

Philipp Capetian

QWEN Q6_K runs through smoothly without error codes but leaves me with a black picture only...

Taiga

wan2.2_t2v_high_noise_14B_fp8_scaled is not found by your downloader. this is the one with the tan lizard. Also, I wish the icons would also apply the prompt that accomplished the image; just as a starter and to confirm that it's working. it puts in the model used, but the user is immediately stuck with not being able to try the preset because the model isn't 'there'. At that point, a not-so-handy message pops up saying, 'model not found'. and you're stuck trying to find it locally, than searching online for it... snag, snag, snag.

aicreativeperson

Similar to the two comments above, I had issues with the download on the Gradio tool (which normally works brilliantly). No worries, I downloaded one from Huggingface/Quantstack. The other issue is the image going black. Removing Sage Attention from the ComfyUI Backend ExtraArgs field seems to get it working. Similar to the SageAttention issue with the CausVid I2V Lora. Not your fault but just so you know. Thanks for trying to get it to us so quickly though!

Kevizz Men Of The World Art Adult/18+

Yes removing Sage Attention from backend helped in downloading without errors in Swarmui. I still have to generate an image to see if it works.

SECourses

Yep not working yet I just published tutorial and explained

SECourses

I will check and fix. You downloaded Wan 2.2 bundle?

Joel Maynard

can we train lora or dreambooth with Qwen?

David

I get this error: All available backends failed to load the model 'E:/SwarmUI/SwarmUI/Models/diffusion_models/Qwen_Image_Q8_0.gguf'. Possible reason: ComfyUI execution error: Unexpected architecture type in GGUF file: 'qwen_image' I updated ComfyUI and SwarmUI as I do every time.

Serkan köksal

Hocam ben de hem comfyui'da hem de swarm'da Qwen çalıştırmak istediğimde "torch uyşmazlığı" hatası alıyorum. güncellemeleri yaptım, gguf ve safetensors sürümlerini denedim. sonuç aynı...

Taiga

I have several Wan files; not that one. I just wanted to try the preset.

SECourses

bizim yöntem ile yükle ve başka yükleme yapma. ekran kartın neydi?

SECourses

not up to date. if you cant fix just make a fresh install. but latest installer file updates GGUF too

SECourses

yes coming soon hopefully kohya working on it

SECourses

just change models with your own models simple as that

SECourses

i just checked latest version of the app and this is in the app and also in the repo : Wan 2.2 T2V High Noise 14B FP8 Scaled ; {"name": "Wan 2.2 T2V High Noise 14B FP8 Scaled", "repo_id": "MonsterMMORPG/Wan_GGUF", "filename_in_repo": "wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors", "save_filename": "wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors"},

David

Fresh install of comfyUI or SwarmUI?

SECourses

comfyui. this is comfyui error it cant understand. also run swarmui updater too dont forget

David

I already run the SwarmUI update. Ok, when I'll be home I will try to reinstall comfyUI and download again the gguf model. I will let you know. Just one question more, I use to update SwarmUI from inside the app too, it do the same of your update bat file?

SECourses

update bat file should be sufficient. make sure to close app update and start

David

I will test it, when I see from the app and update I don't click on it, I close, I execute your update bat file and come back in SwarmUI to see if it don't show the update anymore 👍🏽

David

Install again ComfyUI fixed the error, thank you. And yes, using your Windows_Update_SwarmUI.bat is the same to press the "Update and Restart Server" button from the webapp. Thanks boss :D

SECourses

you are welcome glad you solved

Artem Servator

Qwen Image training in Kohya Musubi Tuner already available

SECourses

yep hopefully next week i will work on that