CaffeineMix版本v1.0_VAE (ID: 154697)

A checkpoint merge I created a while ago.

I had forgotten about this model until today when I decided to give it a try. It gives pretty good results so I decided to share my work here for all the AI lovers

Version V2.0 :

This version improves upon V1.0

• the model is now more versatile, it can do both semi-realistic and realistic images

• It follows your prompt more closely

• improves the hands

• faces should be now more realistic

• the model now doesn't do Asian faces by default

Prompting Tips

• less is more, use small and simple prompts.

• Use small negative prompts.

• avoid using tags like "1girl" , it makes the images follow anime/render style.

• don't describe things too much, it can make it difficult for the model to do it properly.

• 20 steps seems to give the best output, u can try higher if u need.

• CFG scale of 5-7 seems to be the ideal range, personally 6 seems to be the sweet spot.

• use adetailer for better face.

• use DPM family of samplers ( u can decide which one u want to use but I personally like DPM ++ 2M SDE).

link to full image

描述:

A checkpoint merge I created a while ago.

I had forgotten about this model until today when I decided to give it a try. It gives pretty good results so I decided to share my work here for all the AI lovers

训练词语: