NoobToIllustriousVpred v1.1
NoobToIllustriousV_v1.1 has the same recipe as NoobToIllustriousV_v0.1, but instead of Illustrious-XL 0.1 I used Illustrious-XL 1.1
NoobToIllustriousVpred v1.0
NoobToIllustriousV_v1.0 has the same recipe as NoobToIllustriousV_v0.1, but instead of Illustrious-XL 0.1 I used Illustrious-XL 1.0
NoobToIllustriousVpred v0.1
Purpose of this merge
The purpose of this merge is to convert Illustrious 0.1 to the Vpred version. I did this merge simply because I was curious what would happen if I tried it. It's not the best quality model compared to other modern models, and it's more of a proof that you can convert an Illustrious model to a Vpred variant and get something usable. Maybe you'll find a use for this model, for example, to create merges.
This is a Vpred model, so you will need the correct ComfyUI workflow for this. You can simply drag and drop the example images into ComfyUI to get the workflow.
Recommended negative prompt
very displeasing, displeasing, (low quality, worst quality, jpeg artifacts, vhs artifacts, aliasing, blurry, distortion:1.4), (bad anatomy, bad hands:1.4), (muted color:1.2)Recommended CFG: 7-15
I also recommend using artist tags, as the default model style tends to add VHS-like artifacts to images
Recipe of the merge
1 Create temporary model
TempModel = NoobAI-XL Vpred 1.0 + Illustrious-XL 0.1 - NoobAI-XL Epsilon 1.0
2 Merge Stabilizer nbep11 v0.138 with strength 0.5
3 Merge Stabilizer illus v1.23 with strength 0.5
4 Apply CyberFix with CyberRealistic XL v5 to enhance anatomy knowledge of the model
If you want to know more about what CyberFix is, you can read about it here
https://civarchive.com/models/913998/noobaicyberfix
Logic behind this merge
I was just wondering what would happen if I turned Illustrious into a Vpred variant by adding Illustrious weights to NoobAi Vpred and subtracting NoobAi Epsilon weights. I expected to get just a Vpred variant of Illustrious with the usual problems of Vpred models, such as very strong saturation. But to my surprise, the resulting model had exactly the opposite problems: the model generated images with very muted colors and a lack of detail, even with high CFG. To fix this, I merged two versions of the illustriousXL/NoobAI Stabilizer lora. After that, model became more or less usable. Then I applied CyberFix to improve anatomy knowledge
Big thanks to the creators of these models!
Description
NoobToIllustriousV_v1.1 has the same recipe as NoobToIllustriousV_v0.1, but instead of Illustrious-XL 0.1 I used Illustrious-XL 1.1
FAQ
Comments (5)
How did you merge?
I did every step of merging in ComfyUI. You can read about merging models in ComfyUI here
https://comfyanonymous.github.io/ComfyUI_examples/model_merging/
@MindInTheDigits Hi, can you share your workflow?
@MindInTheDigits I want to copy this model but cannot, can you upload the workflow?
@jmeswin10 Well, I just did a simple "Add Difference" merge. You can use this workflow
https://civitai.com/models/991574/noobaicyberfixmergeworkflow
Just replace NoobAI-Epsilon075 with NoobAi-Vpred1.0, cyberrealisticXL_v31 with Illustrious and stable-diffusion-xl-base-1.0 with NoobAI-Epsilon1.0
I didn't merge all of this in one workflow because I don't have enough RAM for it. First I converted Illustrious to a Vpred variant using Add Difference merge with NoobAI-Vpred1.0, after that I saved this temporary model to disk. Then I merged the Stabilizer loras with the model and then saved to disk. After that I did CyberFixMerge.
If you don't know how to merge loras in ComfyUI, you can just follow the instructions here. It's very simple
https://comfyanonymous.github.io/ComfyUI_examples/model_merging/
Also, here is instruction on how to do CyberFixMerge
https://civitai.com/models/913998/noobaicyberfix



