Now What?

Now that my lip sync tests are done, I will be continuing on with more story-driven content. The lip sync is not perfect, but I think it’s good enough to proceed. TLDR: Going forward, I’m planning on simultaneously working on both longer SFW animations with closed-source models and longer NSFW animations with open-source models.

There are several options for where to go from here. I’m going to be very transparent because you guys have chosen to support me (thank you!) and I feel like I owe it to you.

Option 1: Use Runway Gen-3 Alpha to create high-quality SFW animations with lip sync. (Proof of concept here)

I know a lot of people (myself included) have been blown away by the realism of this model, but there are too many drawbacks to go with this option. These include high cost ($95/month), the inability to steer things with starting and ending key frames, and very high censorship. I won’t be moving forward with this option.

Option 2: Use Luma or Kling to create high-quality SFW animations with lip sync (Proof of concept here).

I’m still going back and forth on whether I should use Luma or Kling but both models work more-or-less the same. Drawbacks include cost, some weirdness across transitions between the 5-second clips, and censorship. However, I can steer things quite well with these models which means I can tell a story with them. Moreover, the feedback I have gotten is that quality is more important than NSFW elements to a lot of you.

I will be moving forward with this option. My next project (if I can make it work) is a male-to-female animation of a man transforming while sleeping. As he lays in his bed, he dreams of growing up as a girl, his body changing to reflect his shifting memories.

Option 3: Use AnimateDiff to create lower-quality NSFW animations with lip sync (Proof of concept here).

This option is free (open-source) and highly customizable through different models/LORAs, but the drawbacks are that it is lower quality and takes more time. Still, I don’t want to give up on using open-source AI. It’s probably worth taking some time to explain why.

First, I can generate anything I want with open-source models, and I mean that literally. If the model can’t generate it currently, I can finetune the model on any image or video I can get my hands on. With Luma or Kling, I’m limited by what the creators decided to train into the model. Second, open-source is forever. Luma could go bankrupt tomorrow and their model would never be seen again. If the model lives on my hard drive, it’s not going anywhere. Finally, I believe open-source is the most likely path to full NSFW, photorealistic animations. Tech companies are reluctant to allow NSFW generations on their websites. We are where we are with NSFW AI images because the censored models were released as open-source, and then random community members finetuned them on terabytes of porn. My money is on AI video following the same trajectory.

I will be moving forward with this option. My next project (if I can make it work) is a female-to-female transformation of a camgirl who incorrectly uses the tags #hairy #bigboobs #asian when none of those tags apply to her. The camsite she’s on transforms her to match the tags.