I am half the way in writing act 2, and I haven't done explored anything else regarding the AI tools I'll have to use to create the images for the Visual Novel. 

Unsurprisingly there are now new things available that I need to look at, to see is they would be useful.

I probably lacked the incentive to go looking, because I wasn't sure what I was going to need until I had written enough of the second act.

I had already identified how crucial Loras would be for characters, and for any returning character, I would surely use Loras for consistency of the faces.

There is also controlnet. I am going to use that for action oriented images, so I will use it sparingly, as there is not a lot of action planned.  Controlnet in action can be seen below, I only had to feed it a stock image, in a pinch I can draw/photoshop something and then feed it to controlnet to get what I want.

Getting suits, uniforms or dresses consistently is not really that important in these scenes. Choosing images with a backlight will help too in these situations.

While these scenes are going to be used as accents, the regular narrative will continue to be conversations, which leads to the need of consistent faces. When I began, before I decided that I would need Lora training, I had already looked at a reference only model or a 1 image controlnet, but the results weren't entirely satisfactory

Now I found out that there is another solution to adjust the model based on a single image IPAdapter.

In the following example, you can see it in action. On the left, the base image (This is Mary, one of the characters for Act II). On the right the result of applying IPAdapter to the base image and using a similar prompt but changing "Blue dress" "In a kitchen" and "Frowning".

I didn't get this result at the first attempt, but you can see that it works. Maybe I can use this first to move the character to a different setting then use face detailer + Lora to get expressions and a stronger resemblance to the original, and also to generate several images that I can use for Lora training. Another tool for the toolbox.