DuxVector
patreon
May 2025 Monthly Report, V4.5 is Live! Requests for V4.5 are available now!
🕑 Added 2025-06-01 02:11:42 +0000 UTCComplimentary little Luci-chan comic at the bottom as usual!
Kept ya waiting huh?
TL;DR (SHORT VERSION) -
Requests open!
V4.1 discontinued. Say goodbye~ 👋
V3 support dropped. Will still exist for requests, but strongly dissuade people from using it. V4.5 at this point has 50+ new poses, much more in-depth facial control, tons of bugfixes and on average gens almost twice the number of images. V4.5 feels like a true spiritual successor to V3, unlike V4 which was a questionable sidegrade at the time. Content cycle from now on will be in V4.5 only (until the next hot fresh update comes along)
Content remains Black->Black->White for now.
V4.5 performed very well, has minor issues, but will be fixed.
Now the rambling begins.
The V4.5 eagle has landed. I repeat, V4.5 eagle has landed. If you're reading this that means the sets have launched successfully and the posts were scheduled to all drop together.
The entire system was prepared ahead of the V4.5 launch to fully integrate with it. I can safely say on my end, everything was successful.
The NSFW AI Creators who are genning full V4.5 sets from Day 2 gotta be a handful honestly. You're at the cutting edge of it all ladies and gentlemen~
At this point, you must've seen how V4.5 has performed, and while I understand it's a significant departure from how V4.1 looked, it now feels a lot more like a spiritual successor to V3 than what the "V4 incident" was.
Funny enough 
NAI admitted their mistakes and that V4 had issues. A stark contrast from the day 1 smug and condescending tone of "bro just prompt better bro".
Now V4.5 doesn't come without issues so let me go into some quick Pros and Cons -
Cons
Oversatured colors - without a doubt everyone's #1 complaint so far from what I've seen on social media. They turned the knob all the way to the other side in fear of doing another V4 incident where everything looks muted and dull. The colors are a bit over the top, but I've already implemented techniques even in the first Demo Sets to reduce the ridiculous color saturation. Some people like it, myself included, to have a bit more color than normal, but I understand that a middle ground has to be struck for everyone to be happy.
Composition and framing - this isn't that big of an issue, but V4.5 really doesn't like wasted space. It'll move the camera as close as possible to fit the exact amount of things that need to be in the scene. While nice and intelligent, it can make images feel a bit claustrophobic, up close and very personal. It has affected a small portion of my poses, but I'll finetune them pretty soon. (For example doggystyle is generating only Girl Leftside and Male Rightside right now)
That's it. No really, that's pretty much it lol.
Pros
Extremely up to date database - I haven't ran tests, but it's legitimatelly trained up to probably May 2025.
Proportions and artist mixing - things feel meaty again. V4.1 was an intelligent model as well, but it kept the proportions rather slim and overly reasonable. We like it unreasonable here. Artist mixing is responsive at least for me, which means that slowly but surely any visual aesthetic issues that need to be addressed can be finetuned quickly.
Eye quality - with the launched Vivian set, I intentionally picked a character with complex eyes. She has these vampiric-like, vertical slit style pupils with a slight blue backlight to the red iris. Even in mid-range images, the design was correct. The color-in-color backlight would mostly be only seen in close-ups, but I mean come on, still pretty good tbh.
Color and volume - for everything really. The backgrounds are colorful, the clothing is colorful, skintones are fleshed out well, facial blush is colorful, etc. It isn't holding back.
Symbol-shaped pupils - heart-shaped eyes are very consistent. Again, the closer they are the beter, but even in mid-range it did fine. As well with the Vivian set, I tested whether there'd be a clash between abnormal pupil shape and abnormal eye shape tags. Nope, they worked fine.
Intelligent model - it's handling complex scenes very well. Frankly, not as good as V4.1, but that's probably because the prompts are written for V4.1, instead of being written for V4.5. Skill issue. Will slowly get to rewriting them all.
Very good baseline - the baseline images are extremely high quality, meaning inpainting is something that will only be done to address structural flaws in an image rather than amplify the quality of the images. In V3 there was a pretty stark difference between the focus inpainted and non-focus inpainted images which I never liked, but was a necessary sacrifice. Translation - more images per set.
Sweat and messiness - tags are getting very responsive. Everything from saliva, to all other "fluids" are now consistently "showering" the scenes. This was a main gripe people had with V4 and despite being able to slightly increase it with V4.1, it was obvious the model was very reserved when it came to that. Well V4.5 doesn't have that problem. The faucets are flowing as they say (who says that?). I'll look into potentially messier hair and messier makeup pretty soon. Don't worry, for those who don't like that, it won't come at the cost of anything. It will all be an added bonus to the set.
Expressive emotions - V4.1 struggled conveying emotions all that well. It wasn't bad, but more nuanced emotions like nervousness and shyness were a bit hard to convey. V4.5 doesn't have that issue.
Mouth control - may sound weird but it was actually surprisingly difficult to properly control the mouth in V4.1, how open, how wavy, when teeth show up, etc. People probably noticed a prevailing :o (really simple slight open "O" shaped mouth) that looked kinda basic and uninspired. Yeah, that's gone now.
Fingernails and nail polish - it depends on the character a bit, but the Vivian set exhibited a ton of really well detailed distinct shaped fingernails with colored nail polish as well. Just another strength of the model.
Lipstick - I keep lipstick mostly freeform, but in both Demo Sets it came ou well and has proper application. Things such as kiss marks are looking good too.
Makeup - you guys know I love my makeup. Luckily this time it's very consistent and reasonable. V3 would sometimes just bomb a character with super heavy eyeshadow which felt unnatural. V4.5 sprinkles just the correct amount.
There's a bunch of other stuff underneath the hood that's good, but for now I'll keep the yapping low.
Good and Bad? Depends on your perspective!
The Males are Greek Gods - they are literally built like brickhouses. God damn, everyone's a Mister Olympia winner here? Though I personally don't dislike that, I'll maybe tone them down around ~20% (if I can lol). The size difference does look great though. V3 had it as well, but when larger framed big booba anime girls came along, sometimes the guys just couldn't match up to the size. Now there's no problem. Also their other "sizes" have seen a substantial increase as well.
Veiny - things are very veiny again. That's how it was in V3. V4.1 tried to keep things more normal looking but we're back to explosive blood pressure territory again. Again though, personal preference of mine.
Inpainting is delayed - technical stuff that doesn't really matter all too much to you guys. But they basically rushed the model out the door as quick as they could and the inpainting tool (the thing that lets me fix errors in the images and amplify certain things like eye clarity and small details) was left behind a bit. What this means is that some details on the images can look even better, just gotta wait a bit. It technically still works, but uses the previous models functionality which can cause discrepancies in consistent output.
Motion is a bit too aggressive at times - while I enjoy having a sense of motion in an image, it can be a bit too much when there's motion blur, motion lines, after-images and impact morphing (for example when the males hips hit the females butt it warps it a bit). I'll try to strike a balance where there's a proper feeling of motion that doesn't come at the cost of visual clarity.
That's it for now.
It's going to be an ongoing process. Bits and pieces will be improved each week. This is going to be a lot easier than improving anything on V3 considering everything felt like pulling teeth back then. I've also rewritten a lot of my code to be much more streamlined, much easier to scale up, much easier to debug and more!
If anyone has any specific questions they're more than welcome to ask in this post, or send me a DM on Patreon, or reach out to me on Discord.
Hey hey Luci-chan~! Look! You can have proper comics now!
I uh....gotta go. Lotta work to be done on the new model haha!
Comments
CO_TJ
Thanks! I'll look into this issue. Can you DM me the error message you're getting exactly?
kuroitanuki
Can't download due to capacity issues with MEGA
