It's been a fairly unexciting week, not a huge amount to report. I completed everything except the SFX and music for Alice's event and I'm about half way through doing Emelia's. I should hopefully have that done this weekend.

The last week has mostly been a tale of frustration over what can and can't be done with video generators. They have really weird and overly cautious limitations, after about 10 gen attempts (which all cost credits from the monthly balance I get from a subscription) just trying to get someone to playfully slap/pat/tap (and a full thesaurus of alternative words and descriptions for this action) another person I had to give up. Just now I gave up on a scene where the MC has his back to the camera with only tiny part of his face visible. The animator tools kept making him turn his face to make it visible and the faces they come up with... horrendous. The first time it did it, I realised I was going to have to specify the general look of the MC for when it inevitably turned his head. It doesn't matter if you specify him not to, or that he is looking somewhere else. If the starting frame is a certain way it WILL do it and there is nothing you can do to stop it. Somehow multiple different generators turned "Caucasian, handsome, clean-shaven British male with short straight dark hair" into a someone that looked like they had an auburn coloured afro, facial features that looked like botched plastic surgery, a weird afro beard and heavy makeup on. Multiple times. So that scene has had to be redesigned as well. It can be quite demoralising when you run up against these brick walls, but I just have to tell myself that human imagination is a pretty powerful thing and if I can't animate it, it isn't the end of the world. I just make sure the description of it and the sprites and scene composition are good enough. Anyway that's my therapy rant for this week.

Speaking of scene compositions. When I first started this I mostly used background scenes I had generated using standard SDXL checkpoints. Which does the job, just not particularly well - especially if people are in the scene. Which is why some of the early scenes in Luminaria have quite botched faces in the background characters. I later switched to Flux Pro, which is where nearly all the backgrounds from about half way through Chapter 1 onwards came from. For the last update I also started using Nano Banana. Flux does great stuff, but for general backgrounds it tends to produce very similar overall compositions and it wasn't working for the rainforest scenes because it was adding in well worn and easily traversable paths into most generation attempts. I tried a few different ones and Nano Banana was good at not doing that and creating more varied scenes, so I started using that too. For the Emelia event I needed to do more scenes in Luminaria with people in the background. I tried out SeeDream 4 and was pleasantly surprised as the images were pretty good, it handled concepts like fusions of modern and and old fashioned clothing well and the background faces were much better. Still problematic but fixable with face improving upscaling tools. It's also very good at following instructions. On my patreon page you can find some of the new backgrounds (you only need the free membership to see these sorts of images), including the redesign of the street where Vivian's shop is located. It's a little bit basic in terms of colour schemes, but I think for future stuff I can probably get much more intricate designs out of it. Flux will do wording and signs, but nothing like as consistently as this. I don't think it would allow me to specify the signs and parameters for 3 different shops with signs and actually do them, whereas SeeDream did. So that's a nice little find. Unfortunately its 3x as expensive as flux to use, but I need less generations to get what I want so there is a trade off there.

I think everything is looking good for the early access release to be out in the usual 1 month window (first week of November). Once these events are done I'm going to re-do Vivian's scenes in Chapter 1, transfer more character profiles over to the new format and work on making the end of content messages and mission or event availability systems and codex screens much more user intuitive. I've seen some feedback that suggests it's not clear what you need to do at that potential end of content screen, so it needs some work.