Development Update 16/11/2026
1.04R is nearly finished. Just a few finishing touches but it is now basically all done. The last week involved tidying up Lilly's event and completely putting together all the assets for Sarah's event and getting them scripted in.
1.04R is nearly finished. Just a few finishing touches but it is now basically all done. The last week involved tidying up Lilly's event and completely putting together all the assets for Sarah's event and getting them scripted in.
The public release of 1.031R will go out tonight, sometime around 11-12 UK time. I expect the EA version of 1.04R to go out later this week, probably Thursday.
I don't have much to write-up, Lilly & Sarah's events have all been pretty smooth sailing. I feel like I've gotten a much better grip of combining Illustrious and SeeDream to get the results I want. I've barely had to use LTX2 for any of the animations in this last update. Which i find is saving me a fair bit of time. LTX uses start images or start and end images to make the animations. That tends to mean I have to create a lot of start images for variations, to create variation in the animations. But SeeDance, Hailuo, Kling and WAN3 all have the option to use reference images instead. There are good and bad sides to reference images. A good side is that the character doesn't have to start or end, with their face fully visible and clear on screen, in order to show that character doing anything. The bad side is that the AI can end up doing stuff you didn't want, or incorporating the references in unpredictable ways. SeeDance and Kling are expensive (there is a new version 4 of Kling out, it's good but still only SFW and it costs the same as SeeDance to use). Hailuo and WAN3 are mid-price range, but the image quality and the breadth of animations they do aren't quite as good (but generally good enough). So when the AI goes off on one it's painful because the budget is limited. Whereas with start and end frames you can force the AI to do what you want. The difficulty is that you need to have both the start and the end image and they need to broadly be in line with each other, else you see a lot of warping or jump cuts because the animator can't make sense of them).
I've learnt how to get more out of SeeDream image generations. SeeDream tends to try to make people too big when it adds them to the scene. You can rework the iamge and ask it to shrink the person, but it's hit and miss. However, I've worked out how to get it to put the person in at the right sort of position and scale, using a crude version of the scene made with a sprite. There is an example of this in the screenshots, as well as a few other shots from the 1.04R content.




