PMV Director - New tool to create PMVs

[CantoFan123]

Wowowee!! This is amazing! I it always made sense to me that using AI to edit PMV makes so much sense. And you've done it! I don't have time in my life right now to try this out, but some day soon I would love to and I will be keeping an eye on this development!! Thank you!!

I have always wanted to see more PMVs with x-art, which is my all time favorite studio and is rarely PMVed I think because they aggressively enforce copyright. Being able to create my own would be huge!!

Also I like to watch pmvs on my home theater system which has a nice set of speakers and many PMVs use super compressed versions of the songs, so being able to select my own high quality tracks would also be amazing.

I think AI is mostly overhyped but in the realm of porn it is FANTASTIC! I think the same of VR! Altman and Zuckerberg are just teaming up to put the ultimate gooning tech into every household!

THANK YOU!

[suzyjamestv]

Hi All,

Thw updates and changes just keep getting better and better

SuzyTV

[MrLobo]

Hey everyone!

Just wanted to give you an update on that post from almost a month ago where I mentioned diving into new modes beyond just PMVs. Well, the "mistress" mode is still in development! It's taking me longer than I initially thought, but honestly, I am loving what it’s capable of doing with all this video analysis data. Of course, there’s still a lot of tweaking and refinement to do. The next update will use the new models based on Qwen 3.5 and Gemma 4. These are performing so much better and they're faster while taking up less space!

I hope to roll out this new version very soon!

[Minuit]

This sounds great !

[MrLobo]

I apologize for the long wait, but I've been adding significantly more features than I originally planned. This has taken longer than expected, but I am thrilled to announce a new version of PMVDirector!!

Here's what's new:

Spoiler: show

Bug fixes
- Several bugs have been squashed.

Export MP4 using GPU
- When exporting as an MP4, if a GPU is detected, it will be utilized instead of the CPU, which results in faster creation of the video files.

Goon Captions & Goon Voices
- You can now add animated goon captions and custom voice-overs to your videos, whether they're newly generated PMVs or any external video you throw in.

Example of this new goon feature (added goon captions and goon voices to the round 2 of my cock hero, 'Why Are High Heels So F#King Sexy'): https://pixeldrain.com/u/9WnpHVm1

Easier Configuration
- Setting up the tool is now simpler. A new "config" option in the GUI lets you verify the correct configuration
Example of bad config:

Example of good config:


The Mistress Mode
The all-new Mistress mode turns your videos into an interactive experience. Here's what it brings to the table:

- Your videos, your Mistress — she uses your own video library (vlib) to generate targeted teasers and challenges based on the videos you already have. Your collection becomes the playground.
- Full session generation — create complete sessions with photos, text, voice, and videos. The Mistress picks challenges, generates the content, and assembles everything into a cohesive experience.
- Evolving challenges — every session feels different. The challenge system learns from what you've played, so no two sessions are the same.
- Talk to your AI Mistress — have a real conversation with her. She responds in character, adapting to your personality, your story, and your preferences.
- Photo teasing — the Mistress pulls real photos and analyzes them to tease you about exactly what's hottest in each image, referencing your personal kinks and history together.
- Gallery browsing — she offers you photo galleries with a price. Accept, decline, or pass — she reacts in character to every choice you make.
- Powered by pornpics.com for photo fetching — the Mistress scours the web to find the perfect images for each interaction.
Note: The interactive chat feature is experimental and may evolve.

Example of how to create a new Game:


Here you can see this demo session exported as a video: https://pixeldrain.com/u/HCBLs3Q2

Pics of the interactive chat:





Voice Cloning
- Connect any TTS service that follows the OpenAI spec
- Want a custom voice? Use the voice cloning feature: select a reference audio file, and you can use the cloned voice. Perfect for creating personalized narration. You need an external TTS and connect it in the config.toml, I recommend using this one https://github.com/maemreyo/omnivoice-server (Docker installation is quite simple)

For this new version, it is necessary to download new AI models and update the llama.cpp version to one of the latest versions. There are also new variables in the config.toml file, so you will need to migrate your current configuration to the new setup.

I have updated the original post with the new download links, and I've also updated all the information on how to set everything up. I hope I haven't forgotten anything! If you have any questions, please ask them here or send me a private message and I'll help out however I can!

[Queenbee15]

Hi Mr Lobo,
First of all thanks for your amazing tool!, myself and my boyfriend have made quite a few vids with this.
I do have a question, would you be able to add like a Girls mode?
I would love to be able to create like a Dildo Hero type PMV with this.
For example the AI keeps referring to me as a guy etc. Same with the Mistress mode.

thank you so much!

[MrLobo]

I'm glad you're using the tool
Right now, the gender is defined in the mistress prompts and it defaults to assuming you are a male, but I can definitely make it configurable. Thanks for the suggestion; it's something I will keep in mind for the next version.

[anjinsan2020]

Hey there, I've been using the tool extensively to get familiar with it and it's really an incredibly well though out interface. I was wondering if there would be a way to bulk edit text descriptions for segmented clips in a video. Something that could use a CTRL+F style search where I can remove incorrect terms assigned to clips in bulk. An example would be where Llama.cpp identifies "nipples" in almost every clip even though there are none obviously visible. Right now, I have to go into each clip's text box and edit them manually which is definitely a labor of love but one I can do without. For me that seems to be the main concern as it takes a large amount of time to comb through the scenes one by one. Of course being able to train the model would be even better as it would get much better at processing images and scenes and deciphiring correct attributes. Maybe we could collaborate on this as I have some experience training AI models. Let me know what you think and thanks for all the hard work you've put in so far!

[nostradamus]

Hi,
After installation all checked out green in config screen I ran the software, created library, added a folder and get this message when I click start processing. Not sure where the log file is and how to fix it.
Thanks for the effort put into creating this software.

errorpmv1.jpg (291.83 KiB) Viewed 1894 times

config.jpg (110.31 KiB) Viewed 1894 times

[nostradamus]

Hi,
After installation all checked out green in config screen I ran the software, created library, added a folder and get this message when I click start processing. Not sure where the log file is and how to fix it.
Thanks for the effort put into creating this software.
errorpmv1.jpgconfig.jpg

Mr. Lobo ! It is working now ! Generated a video with beat and voice. As other users mentioned the software doesn't recognize cum shots and anal sex.
Thanks a bunch for this master piece.

[MrLobo]

Hey there, I've been using the tool extensively to get familiar with it and it's really an incredibly well though out interface. I was wondering if there would be a way to bulk edit text descriptions for segmented clips in a video. Something that could use a CTRL+F style search where I can remove incorrect terms assigned to clips in bulk. An example would be where Llama.cpp identifies "nipples" in almost every clip even though there are none obviously visible. Right now, I have to go into each clip's text box and edit them manually which is definitely a labor of love but one I can do without. For me that seems to be the main concern as it takes a large amount of time to comb through the scenes one by one. Of course being able to train the model would be even better as it would get much better at processing images and scenes and deciphiring correct attributes. Maybe we could collaborate on this as I have some experience training AI models. Let me know what you think and thanks for all the hard work you've put in so far!

Thank you so much for the kind words! I'm really glad you're enjoying the interface and taking the time to test things out. I’m currently focused on other features but definitely have this in mind. If you train a model that improves the descriptions, please share it, it would be fantastic to have one specifically tuned for this use case. While the models listed in the installation guide have given me the best results so far, a custom-trained version would be ideal.

Mr. Lobo ! It is working now ! Generated a video with beat and voice. As other users mentioned the software doesn't recognize cum shots and anal sex.
Thanks a bunch for this master piece.

So sorry for the late reply, I’ve been offline for a few days. I’m glad to hear it’s working now! Looking at that error you mentioned initially, it definitely pointed to an issue with the Llama.cpp installation or configuration.

Regarding the lack of recognition for specific acts like cumshots and anal sex: this is because the default model used for frame description is generic and not trained on adult content. If anyone finds a model specifically fine-tuned for this purpose, please share it! It would be amazing to have one available, as switching to that model alone should significantly improve precision without needing any other changes.

By the way, if you increase your processed video library, I highly recommend trying out Mistress Mode, it’s easily my favorite feature. I never get tired of the AI generating custom teases for me

[variationx]

Amazing work!
I know you're not a dev, but do you have a Git repo or can you publish one with the code? This would take-off like crazy for the creator community.

One feature I'd add is a sync beat to stroke / penetration. This would involve re-time or time-stretching (something that is manually painstaking. This however requires some kind of flow analysis and image recognition to determine when hips meet and penis is fully inserted into an orifice. Beat should be able to be on the max penetration or on the max extraction.