Delavestra
patreon
All 5 Zoom Set - Erotic Public Previews
🕑 Added 2025-02-25 06:46:36 +0000 UTCHi all, buckle up. The Zoom set series is the next model/lora/prompting style formula that I've been obsessed with. I test about 5-10 completely new models and lora combinations monthly before finding one that's worthy to share. I've had a bit of a dry spell with finding unique combinations I've liked, but wow was the diligence worth it.
In my attempt to produce high-quality ai artwork at scale that can tell a story and be visually interesting and not just lewd or vulgar, I believe I've found an excellent balance between maintaining my personal style, and allowing for full futa capability with quality and appealing aesthetics (most models cannot do futa at all, fewer can do it well).
I have Female and Futa labeled versions in the full sets which are in their gallery, and available for ZIP download to support me further. It's a challenge to properly organize the eye-candy chaos. I'm very happy with the these sets and I already have more complete and in-process sets based off of this style.
When you're ready, click in and enjoy! No need to keep reading...
For those of you interested in the 'how' I produce these, here's a quick walkthrough which I'll elaborate more on in the future. I'm currently using the UI called FORGE. I run my generations locally on my 3090 but I've ran on H100s and A6000 on Runpod for video and faster image production. Looking to get a 5090 soon so my RnD goes faster!
I'm currently using one of the Pony Models and I utilize between 6-13 LORAs per image, depending on what I'm trying to create. Now, here's where I'm different...
I've created my own prompting scaffolding that utilizes combinatorial generations to allow for bulk image runs when I'm ready for a production set. That's nothing fancy, but the part I'm now refining is the use of Grok3 and custom made tools... Grok3 is uncensored so I now have a custom system prompt I've provided Grok to aid in my combinatorial generations to allow for faster and more efficient prompting structures when creating my images. I'll be sharing a prompt from my Melissa and Ariana set in the coming days so you can see how bulky and cumbersome it can get.
I then use a series of custom-made tools for sorting and labeling my images. I have a program on my pc that automatically creates my folder structures and allows me to 1 click (per image) sort images at scale with the single push of a button. I'm then able to bulk process and upload images to the interwebs. I use make.com to daily post on X (It's still Twitter), Instagram, Deviant Art, and multiple Reddit threads.
I'm continuously trying new AI tools to generate and design the best possible content that's possible. Refining what works and making it better vs trying the new tools that are popping up daily is quite the debacle. I use Klingai for most of my videos and I use live portrait ran locally with a custom API tool for talking avatars. The super high quality videos and talking Avatars are from KlingAI videos paired with Runway's Act One with a driving video of literally me talking, so it's my voice and facial expressions (mo-cap), and W-Okada for my voice changer....
Okay that's enough info for now as a quick summary of what I'm working with. My goal now is to spend more time testing and producing high-quality AI Art for you rather than filtering, sorting, and figuring out how to properly post my generations and determining how to best advertise... This is the most cumbersome part of my process and it slows me down horrendously. But, I'm getting better daily and lots more to come!
For those that are still reading this, I truly appreciate you. Your support drives me to continuously improve my quality and content. Thank you so much! Looking forward to hearing from you soon.
-Del
Comments
Delavestra
Thanks! Yes I'm more technically inclined and do this far more manually than anyone realizes. I don't use paid tools or websites, I run it all myself (Except kling of course). Re voices, any AI voice that's doing tts does not sound good enough for me to want to use. I've tried it, it sucks. Plus, the facial animation taking my face and applying the voice gives a far superior result to any tts and talking avatar... What I really want is a Heygen clone avatar, but their system requires real time safety checks that cannot be circumvented, so I can't make a heygen talking avatar with one of my girls:/ Perhaps in time!
Pombo mix
I'm really happy that you're eager to improve and showcase your work. I had no idea how much effort it took to create these images. I got interested in seeing how this process works, especially since I recently started realizing how complex it is. Your images, at least to me, are close to perfect—not overdone, yet maintaining a very pleasing aesthetic. As for AI voices, I know a little about them. I usually use them in casual videos for dubbing and similar purposes. I used W-Okada for a while, but for dubbing, I started using another tool, which is more specific for pre-defined audio files. You just upload the audio file, and it automatically transforms it into the selected voice model. I think you might know about it, but there's a Discord server where people share all kinds of voice models. Sometimes, people share female voice models that I think could be useful for your work.
