When merging the vPred model with the epsilonPred model, the resulting images didn't turn out as expected. However, by adjusting the block weight values, I managed to retain the TE (Text Encoder) of the vPred model while introducing some noise to create a slightly altered version. Then, I incorporated this modified model into the existing N40NAillous.

The performance doesn't feel like a massive improvement, but the images seem to come out better overall—just a subjective impression, perhaps?

While at it, I also used the newly released NTRMIX4.0 LoRA and arcain-2411 LoRA. Starting from the NoobAI 1.0 model, I redid the entire merging process from scratch, incorporating the vPred model into a newly created merged model instead of the previous N40NAillous.

used LoRA

https://civitai.com/models/960071/better-ntrmix-lora
https://huggingface.co/Bedovyy/arcain

The above sentence was translated using ChatGPT.

1.0

This is a model that incorporates elements from a personal merged model to reduce the skin gloss of NTRMIX4.0.
I took "NA" from the name of the personal merged model, NoobArcaillous, and "N40" from NTRMIX4.0, and combined them.

The personal merged model is a combination of NoobAI-XL 1.0 and Lora, with an additional merge of Arcaillous-XL.

The above sentence was translated using ChatGPT.