The last week has been an interesting one for me. 0.3 went public and feedback was generally positive, but there was also a fair bit of negative. I feel like overall it was OK, but that I made a couple of miss-steps with it. So what did I learn?

Well, I posted the survey about sprite quantity vs quality and the response was overwhelmingly in favour of quality. Which I'm quite pleased about as it gives me a bit more confidence to cut down the numbers. I just need to make sure what is there is all as good as can be. I've also got to train myself to be more discerning in not using all the good sprites. It's surprisingly difficult because when I'm creating and adding in the sprites for a scene, I often have many sprite candidates I can use that look great, and I really want to include them for that reason. But they aren't necessarily adding anything and in some cases, if I let an inconsistency through, they are hurting it. So it's a work in progress as I teach myself to be more ruthless.

With all that in mind, this week the progress has felt a bit slow. I revisited my SDXL checkpoints and generation parameters and spent some time experimenting both with old and new setups. Partly in an effort to maximise consistency and partly because I was having real difficulty getting it to produce images that I know it can do. I eventually identifying that a lighting lora I was using, was severely limiting the range of what I could do with clothing and equipment on characters. So I've kicked that to the side. Which was good because I was getting seriously worried I wasn't going to be able to start layering more equipment on customers in any way that was repeatable for consistency. Having discovered all that, I needed to revisit the generations I had already done for 0.35 and re-do a fair few of them. I've also picked up Luminar Neo from Humble Bundle. Interesting bit of software, it's a photo touch up suite with a bunch of quite interesting AI tools. Infuriatingly a lot of those AI tools will not work if it detects that the image is NSFW, it will refuse to do it. You can still use the local tools, but the really interesting stuff is off-limits for NSFW. I'm glad I only bought it on Humble Bundle, if I'd bought it at it's normal price and then found this out I'd have been livid. However, it's still useful because it allows me to re-light scenes, alter colours, change the sky box, add in mist and fog, denoise stuff etc etc. I've included a couple of examples of backgrounds that I've edited with it to see how much it can do. The non AI tools also allow limited relighting effects, which I've been using a lot as much of Alice's event takes place in the evening at a campfire, so being able to change the lighting on the sprites is nice.

I got Paintstorm via Humble Bundle a few weeks ago and that has been a really good addition. It allows me to nicely fix image issues like the seams around faces. When the base image is generated, the face is then overwritten. That sometimes leaves visible seams in a square around the face. It's particularly noticeable on the neck. But Paintstorm has a whole suite of really intuitive and easy to use tools that allow me to blend out these lines. It also allows me to draw on some nice effects for spells, and for the glowing eyes some characters have. Best of all, it will edit sprites without screwing them up. So many art apps either will remove the alpha layer and add a bad background in, or will add a rough outline all around the image. This helps because I can do all the sprite work in renpy, then go in afterwards and just work my way through the sprites, one at a time, colouring the eyes, removing the seams etc.

Once I finished playing around with SDXL and these new tools, I got to grips with getting sprites and animations into Alice's event and that's now about 80% done. So far, personally, I'm really happy with how Alice's event is coming along. I really enjoyed planning it out and writing it and the sprites and animations I've produced for it are working well. There probably still a few too many going in, but baby steps....

I think I'll have the last 20% done by tomorrow evening and then I'll switch to Emelia's event, which is shorter but will include a sex scene. Those can be quite time consuming to make. There is the method I used to make the images in 0.30 but I've also downloaded a lot of POV lora's to try to natively generate the image in it's entirety. That's sort of working, but the issue with native generation using lora's (rather than feeding it an existing image and telling it to apply the usual character generation parameters I use to create something similar) is that they are trained on porn. And porn stars tend to have a certain look, which isn't very often the natural girl or boy next door look. So these lora's tend to produce stuff that isn't really what I'm after. But there is some potential with them for sure.

I've included some screens on my patreon (free tier) from some of the stuff in Alice's event, including a couple of before and after shots of creating a sprite and editing a background. patreon.com/JubsGames