Jibs Qwen workflow uses a custom-trained Wan VAE to remove the hash/grid lines visible in Qwen outputs.
Here is the GitHub for nodes: https://github.com/spacepxl/ComfyUI-VAE-Utils?tab=readme-ov-file
and the Custom Wan/Qwen VAE model: https://huggingface.co/spacepxl/Wan2.1-VAE-upscale2x/blob/main/Wan2.1_VAE_upscale2x_imageonly_real_v1.safetensors
V6 a Qwen 2512 Version of Jib Mix Qwen.
+ Has better fine details than previous versions.
+ Less of a same face issue.
+ Less nosie artifacts.
- Not quite as good at NSFW as V5. (You can use my NSFW lora to bring it back: https://civarchive.com/models/1943554/jibs-qwen-nudity-fixer-lora)
- Not quite as fine good details as Base Qwen 2512
More European faces and fewer Asian by default.
The Pruned Model nf4 (13.91 GB) is actually a Q5_0.GGUF
V5 -More realistic, less plastic skin by default, with more imperfections.
My Skin detail/imperfections lora can be used to add more or even help remove them if used at a negative weight.
The Version 5 Q5. GGUF (marked nf4)
V4 - Much more natural pretty faces (Much better at Asian faces), less noise, cleaner (slightly less photographic look by default but adding LORAs can make that stronger again)
NEW: The fp8_e5m2 model is twice as fast as the Q5.GGUF on my 3090.
There is a Q5 .GGUF (marked nf4)
Tune for Clownshark dpmpp_3s/Bong_tangent sampler this time instead of Euler_ancestral/Linier_ Quadratic.
V3- Important I have uploaded 2 different versions of this model:
High Noise version (marked fp16) that I think is better at lower steps and single stage workflows (but can sometime show grid/scan lines).
Lower noise version (marked fp32) That makes cleaner/less noisey images on the first gen but is better suited to 2 stage Hi-res fix workflow (this is my preferred method)
There is a Q5 .GGUF (13.91 GB) that is the Higher Noise version
The Q6 .GGUF (15.6 GB) is a lower noise realistic skin version.
Small Q5 .GGUF (13.91) is a lower noise realistic skin version.
V2 - I tried to fix the big bobble heads from the previous versions. It is better but they can still be a bit big sometimes.
The V2 fp8 model listed is now actually a Q8 .GGUF (Thanks to export_tank_harmful for the conversion when mine was broken)
V1 - Adds a more photorealistic/amateur look to Qwen.
The fp8 model listed is now actually a Q6_k .GGUF
and has fewer grid lines than the fp8 (But I still recommend the fp16 with ram offloading, for better quality)
Description
Fixed 2125 workflow (Wan 2x VAE node broke after a ComfyUI update) and used Turbo lora.
FAQ
Comments (10)
Hello. I added a "CLIP text encoder" node in ComfyUI and connected the CLIP from the "Load Checkpoint" node to it but when I click run it says: If the clip is from a checkpoint loader node your checkpoint does not contain a valid clip or text encoder model.
Can someone please tell me where do I get the CLIP file for this model because I can't find it on the model's page or what am I doing wrong?
Thank you!
It is this Qwen 2.5 VL model for the clip: https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/resolve/main/split_files/text_encoders/qwen_2.5_vl_7b_fp8_scaled.safetensors
Having trouble getting the custom vae to work, im getting "TypeError: Cannot handle this data type: (1, 1, 12), |u1"
i used these
https://huggingface.co/spacepxl/Wan2.1-VAE-upscale2x/blob/main/Wan2.1_VAE_upscale2x_imageonly_real_v1.safetensors
https://github.com/spacepxl/ComfyUI-VAE-Utils?tab=readme-ov-file
Load VAE (VAE Utils) Node.
and used other relevant models from here or referanced here.
i can see using the preview that it generates stuff...its just that the vae fails.
i managed to bypass the issue by using the normal qwen image vae, but im curious what that one would do.
Ah yeah a recent ComfyUI update broke those nodes masslevel has made a fork with the fix here: https://github.com/masslevel/ComfyUI-VAE-Utils
I've tried recreating 4 different image style I made with V5 with same promp and parameters and sadly V6 lost a lot of versatily and realism and dont understand certain sci-fi concept and style that was perfect on V5. It is now looking more illustrative then realistic. V6 can look a lot more soft and saturated then V5 too which can look good but is not a style that I like. This looks like it could be another serie since its not a continuation of V4 and V5. But I suppose V7 could be perfect.
I have uploaded a alternative, very early version of v6 here: https://civitai.com/api/download/models/2753478?type=Model&format=SafeTensor&size=full&fp=bf16
that is more like a slightly improved version of V5, if you want to try it?
@J1B Thanks but this model is 39gb and my 5070 TI only has 16gb vram... IT worked ! I got great results, it indeed give me very similar result to v5 and its faster ??? Good thing I have 64bg of ram. Still I dont get why its faster then a 19gb models ! Great results !
Any way to use this with Easy Diffusion?
Yes, looks like they updated it last month to support Qwen models: https://github.com/easydiffusion/easydiffusion/releases
so as long as you are on version 3.0.16 and later it should work.
Please can you do a Chroma1-HD fine-tune as I love that model and it's huge date set and creativity. The creator Lodestone says this regarding the default base chroma hd model: "I haven't done any aesthetic tuning or used post-training stuff like DPO. They are raw, powerful, and designed to be the perfect, neutral starting point for you to fine-tune. We did the heavy lifting so you don't have to."