How can I automate the creation of multi-axis Funscripts?

General
1

Even though I am not a professional programmer, I have put a tremendous amount of effort into automating the generation of multi-axis Funscripts.

I invested heavily in various AI models like Gemini and Claude Opus, relying on “vibe coding” (AI-assisted coding). I also actively researched and applied the latest deep learning and open-source technologies. I went as far as analyzing human skeletal structures, predictive pelvic movements, and anatomical positioning to correlate them with Funscript parameters, even using LLMs to make frame-by-frame corrections.

Through all this trial and error, I managed to build a workflow that generates scripts with about 60% accuracy when I input a target video. However, the sheer variety and complexity of video content make it incredibly challenging to improve beyond that point.

I honestly wonder how much further technology needs to advance before AI can automatically generate multi-axis Funscripts that truly match the quality of hand-crafted ones.

2

You simply don’t have enough training data and a suitable model. Existing models mimic the limbs of the body, not how deep the dick thrust, how the girls hips move in relation to that and angles and depth. IMO, to gain accuracy you need more specific models and (much) more training data.

1 Like
3

If you did manage to get acuracy up though, i think it would really be a multi axis revolution! :slight_smile:

4

More data doesn’t always help.
There is also the issue that different people do or like things differently. Different stroke lengths or curve shapes for the same sections of a video. While the variety can be nice with having separate scripts for training data this is adding ambiguity of output and is going to make for worse results overall.

1 Like
5

If we had a model as powerful as ChatGPT, I think this might not be difficult. However, commercial companies like ChatGPT will never train large models specifically for adult products.