Dionysus 3D
patreon
Experimentation and Practice
🕑 Added 2025-09-08 22:39:31 +0000 UTCHere's a new clip I wanted to share because it is relevant to how my art is evolving. For some of you guys that are interested in my process I thought you might enjoy both watching it and also knowing how it was made. I'm also working with Patreon to be able to post videos directly to the site instead of making you guys download everything.
Prompt for the initial image in Nano Banana (This is a new tool for me), which was the male subject and the two females on either side of him. Fight with censorship to trick it into making the image by changing up the vocabulary I used. For the sake of this post, I counted the iterations. I had to iterate 227 images before I got the one I used, and that's not including the over 500 I did that were censored out.
First, used Nano Banana to add in the female behind and above him, then the pairs of two on the left and right. Again, probably 200 iterations for each of those steps.
Upscaled the image to get it to 2k.
Used Nano Banana again to adjust the light and the various... dimensions... of their body parts.
Moved to video. Used Wan 2.2 (New Tool) to iterate until I got the prompt locked in because it's the cheapest and best model out there for iterating. What I've found is that using the 420p versions of models costs way less, but the animation quality is just as good. It's just the resolution and framerate that suffer. So, you can use those models to iterate until your prompt is right and then only pay for a couple of iterations with the Big Kahuna versions of the models. Had to do about 80 prompt revision iterations to get a version I liked.
Switched to Seedance 1.0 Pro (New Tool). It's hella expensive. Like... hella expensive. But what I've found is nothing is close in terms of continuity of limb movements, anatomy, and camera work without morphing. It's also completely uncensored, so if I have a start image that has boobies in it, it won't freak out and block them like Google Veo 3 or Runway. But before you get too excited, it's dog water at text to video for anything NSFW. Because the model isn't TRAINED on NSFW, it's just uncensored, it doesn't know what to do with genitals or how to interact with them. I should post some outtakes because they're horrifying. There's a whole channel devoted to them in my Discord. I've tried and tried to tame Image to Video by using various descriptor verbs, but it's like talking to a toddler about how to fix an internal combustion engine. However, every time you try to iterate NSFW text to video it's just as expensive, making it completely a gimmick like those games at State Fairs.
Run Seedance 1.0 Pro using the prompt that I dial in using Wan 2.2 at 420p to ensure it translates. Make any necessary adjustments.
Switch to 1080p in Seedance Pro and iterate until I get something that's usable, which if you've done all the above homework usually takes 3-4 iterations.
Run the 1080p clip through a video upscaler to get it to at least 2k resolution and 50fps.
All of that for a 10 second clip.
That's the process. But for those of you who have said to me, "How does your stuff look so much better than everyone else's" that's the answer. Truth be told and all arrogance aside, I haven't found anyone making anything anywhere near the quality I'm putting out, in any capacity not just erotica. For the really crucial shots in my films, I go through that whole process, and my shots are only 5 seconds in length. Butterfly Bay is 22 minutes long. That means it contains 264 5 second clips. Puts it into perspective why they take me so damn long to make, and why your support is so crucial.
That's all for now. If you actually read this, congrats. You're the real MVP.
P.S. Everyone tells me I should do YouTube tutorials. But... like... that would require me to out myself or make videos with AI to make the tutorial, and I already can't keep up with all the projects people are throwing my way. Great problem to have.
