MomBallsAi
patreon
Futa MILF Bath Time - Images
🕑 Added 2025-08-15 20:56:40 +0000 UTCHi everyone! A lot happening the last few days both in life and in the MomBalls world. Mostly downs IRL, but with MomBalls, there is finally a light at the end of the tunnel with all of my Wan 2.2 experiments.
But, for today, I'm releasing an image set I worked on some time ago, riffing off of the Futa Mom Bathrobe project. There are some more images I want to repair from this set, and I'd like to make some videos, but for now, here is what I got!
1 image added
102 images added to Releases\2025\027-Futa MILF Bath Time\ImageSet
Teacher Balls Archive
File change report from 2025-08-13 to 2025-08-15 (since last post)Directories with changes:
- QN-MomBallsVacation: 2 x MP4 (4.48 MB) | 3 x PNG (1.77 MB) | TOTAL: 5 files (6.25 MB)
- U2R-FutaMomBath: 236 x PNG (285.1 MB) | TOTAL: 236 files (285.1 MB)
- V-Natsuyasumi: 39 x MP4 (82.93 MB) | 57 x PNG (64.83 MB) | TOTAL: 96 files (147.75 MB)
TOTAL FILES CHANGED: 337 files (439.1 MB)
In other news:
Warning, long fairly technical explanation below.
I've been wanting to boost my Wan video quality for some time, and now that I have my RTX 3060 12 GB card back from my ex, I can cook on 2 rigs and the added VRAM gives me more flexibility even if it is older architecture. I've done a lot of experimenting with different workflows, models, optimizations, LoRAs and I still have more to do. But, I'm happy to report that I'm finally achieving that quality boost I'd been hoping for. Unfortunately, at the cost of added generation time. One of my biggest pieces of advice to new AI creators is this: experiment and find what works for you. You'll find tons of advice out there, much of it conflicting, but you really need to try things for yourself and find what works for your hardware, for your schedule, and balance those against what you consider quality results. Make a list of the top recommended options out there and start from the least time/labor intensive to the most and stop and lock in your process when you feel the balance is right.
In this case, I really needed a resolution upgrade badly. Generating at 416x608 resolution (and then upscaling to 832x1216) was really holding me back in terms of motion fidelity. The next Wan supported resolution that maintains aspect ratio parity with my source Stable Diffusion images is 624x912 (ideally, I really should change my source image resolution to one that is better supported by Wan at some point, like next time around!). Up to this point I haven't been able to get there on my local hardware (where I do most of my generating) with always hitting the dreaded "allocation on device" error when I exceed the VRAM pool.
I think I finally found the secret sauce for me: using the Wan 2.1 version of lightx2v LoRA weights in both the high noise and low noise Wan 2.2 experts. This allows me to reduce the steps from 20-30 for a good video down to 4-6. I'm still experimenting with what final upscaled resolution I want, but may just go big and keep it doubled to 1248x1824. And also considering if it is time to go to 64 FPS. I found a setup that works on my 12 GB card, but my 8 GB card is struggling. I am going to toy with model quantization, optimizations, etc to see if I can get achievable results. But, the boost in quality for the resolution increase means I can't go back to 416x608, it is just that much of an improvement. I'll post some of my results soon and perhaps a poll to go with it. Of course, resolution/frame rate increases will increase file size, but given how small my files are now, I don't think this is too big a deal.
I also really want to experiment with samplers and schedulers. I've heard great things about the RES4LYF samplers, but most of the videos tend to come out blurry, so something is wrong with how I have more workflow set up. When they do work they are beautiful. They also take twice as long to generate and require custom nodes. Not sure how to get things working with them and there isn't much documentation on on how to get them working in a Wan 2.2 i2v workflow and I can't find any example workflows that do what I want.
And, of course, I really need to wrap up my older Wan 2.1 projects in progress.
Phew, sorry for the long update. Thanks for the support!
