@iamddtla These are not included with COMFYUI by default; you need to install additional nodes.

It really does look better.

@iamddtla You have to install the RES4LYF nodes

Are you asking about Zimage Base instead of Zimage Turbo (which this Checkpoint is based on)? I'd say with your rig you should be perfectly fine. I can run BF16 ZImage Turbo on 8gb VRAM and 16GB RAM. For Base you probably won't have issues with your setup.

What do you mean with "eastern" faces? Chinese, Korean, Indonesian as in more diversity? That could be tough and you may need to search for loras to help with that. Currently the regular checkpoints seem to focus on either Chinese or Japanese women.

@SaburoDio 

I can run Z-Image Turbo. I was asking about the pornmaster checkpoints. Since there are bf16 and fp8 versions, can I run those or are they too heavy?

Sorry, I meant western faces .. my bad. most ZIT faces are already Asian, and the pornmaster samples mostly looks Japanese, so it’s kind of double Asian. I was wondering if it’s possible to generate non-Asian faces with them.

@Airenn I see, yes with your RIG, absolutely no problem to run them. Pmaster does have a focus on Asian characters indeed, but their checkpoint defintely allows western faces and imo makes great ones as well.

I thought about starting reviews to give some guidance over the many many different checkpoints that are available as I see similar questions pipping up all around Civit (Can it do Asian/Western/Males etc etc).

Also realized there are quite a few misconceptions and more so missing info on hardware requirements.

Maybe I will start with these checkpoints and work my way through them all lol

@Airenn oh and just to mention this in general: Some checkpoints accept tags about ethnicity (e.g. "Swedish"). However in most cases you will have to manually tell the model more details regarding skin tone, hair color and other facial features. That is a trial and error process with most models.

Hey, I have the same machine as you. I'm guessing that you also likely have a 9900X CPU if yours is a new build? Regardless, I would suggest installing Pinokio and then Forge Neo. That way you can have a simple foundation to generate images and learn ComfyUI as a side project vs nonstop troubleshooting and asking questions (which is not a bad thing). It is the least technical setup there is, I always suggest it to new users. I was new once, too. Comfyui is no walk in the park BUT it is very good to learn. You can even train LoRAs locally on your build btw.

Does it works with character loras trained on z image turbo?

@iamddtla Thanks

@iamddtla Is there a way to run the double checkpoint workflow with 8GB RAM? Seems to just OOM even if I run the double fp8 checkpoints.

@iamddtla Yes! Maybe I just got lucky, but the ones I tested after training on "z-image De Turbo" which uses these files https://huggingface.co/Tongyi-MAI/Z-Image-Turbo and https://huggingface.co/ostris/Z-Image-De-Turbo/blob/main/z_image_de_turbo_v1_bf16.safetensors

The LoRas that are created work perfectly with your model!

Yeah 100% same here and same acrosss the board, I use the default Config settings Z-Image De Turbo (De Distilled) in One Trainer, thecharacter loras work flawlessly in any set up, including yourr model here of course! .... There are a few that look better in your Pornmaster model than anywhere else actually, the skin details and color defenition really shine through!

I only use One Trainer now for Z-Image loras for me much higher quality than Ai-Toolkit. Many more advanced options also in OT also.

Do you mean that you used this as the base model? I have to have the entire z turbo repo locally (.safetensors won't work in Onetrainer) as the base model and then the De-turbo is my override transformer.

@mustardtiger1776427 The default OneTrainer config that ronikrush mentioned above downloads the repo and file I listed and trains on those, so yes I think what you were saying is right. Then the lora that is created works with this model perfectly.