Sure I can help you with that, I myself recently started to train my LoRa's less than a week ago so I will give you my guide on how I do it.

Don't worry about the fact that your manwha is not NSFW, When generating its the prompts that will make the SFW become NSFW.

Ok so here is how I do :

1- I use an app called "Tachimanga" (IOS only) to collect my data, I can then read the manwha casually and at the same time collect the data.

2- I screenshot panels where I can see A: The face, B: Upper body, C: Full body, priority on faces obviously and if you have the luxury add different angles (sides/back view, from above/below).

Side note: Quality over Quantity, avoid blurry images or images that you crop so much that it becomes blurry.

3- I use a tool named "Anima-TrainFlow" created by ThetaCursed (on GitHub, its free), you can follow the guide on his page and continue. Then inside this tool I start by Bucketing my dataset (min side: 512 & Max side: 768 recommended or 1024 my choice).

4- Ok now the real fun begins, you will have to crop your images, you want to focus on the character as I said in the 2nd step.

Funfact: I find it easier to do this part with SFW manhwa instead of NSFW manwha because yes you have sex panels, but that means a huge presence of males in the images wich you will have to try to remove of your dataset.

5- The second best part is the Captionning of all your dataset, for this I use the CivitAI trainer just to get my captions, then i check if everything i good for this i just caption the image with what I see in the image (character tags, clothing tags, camera position tags, looking at viewer/to the side/away...).

6- All the boring part is done now the training begins, again I use the "Anima-TrainFlow" tool and pray to get a good LoRa, I usually get good results in the 1000-1800 steps. If I get nothing good I either restart it with a different Seed or/and remove some images that corrupt my dataset (blurry, too much cropped or just bad quality images).

7- Try your LoRa's and choose the one that is the best.

8- Enjoy.

@ArtificalEverything Glad I can help a fellow manhwa enjoyer.

I suppose what you mean by tagging is the caption part for the dataset, in this I don’t put anything nsfw because I focus on the character appearance.

Its mostly the model (exemple: anima base v1) you use that helps with nsfw when your manhwa LoRa is only sfw for exemple my LoRa "Han Yena" is purely sfw but by using the right prompt (1boy, hetero, the sex postion you want...) it generate a nsfw image even tho my dataset don’t contain any of it.

For the next sfw LoRa I do I will share my dataset in the LoRa page, you will then see how my images are cropped and see my captions tags I used.

I have to finish a few Pool Party nsfw LoRa’s first before that tho.