PMV Director - New tool to create PMVs

[gooner689]

Suggestion for the Audio tab since the Error "No valid clips found for cut (duration needed)" is still an issue for lenghty PMVS & MUSIC.

Suggestion: The Audio or Video tab analyzes every lenghts of your Imported Videos so it knows what Timeline cuts it needs
= NO MORE "No valid clips found for cut (duration needed)" Errors.

[MrLobo]

Edit - I made some progress with it, I only seem to get the Filename or extension is too long when I'm trying to use longer audio tracks, did several around the 5 minute length with no issues at all, both that I tried that were over 10 minutes produced that same error.

I think I've managed to fix it. Now, for long videos with lots of cuts, the video is exported in multiple batches and then concatenated into a single video at the end, so that error shouldn't happen again.

- regarding the setting roundoff, I now set them directly in config.toml instead of the project, but it looked like some format conversion thing. In the example below the value was set at 0.3 before saving the project, and after reloading it got
Screenshot (1).png

This should also be fixed.

I have updated the links in the first post with the new versions of the binaries.

Got this Error with the new version:

[14:22:54] Error reloading song data: invalid command name ".!ctkframe.!ctktabview.!ctkframe3.!timelinetab.!ctkframe2.!ctkframe2.!ctkframe2.!ctkcanvas2"

Does this only happen with old projects? I changed the Audio tab in the latest version (simplifying it), so when you reprocess the song, it should do so in the new way and the error should not prevent you from continuing.

Suggestion for the Audio tab since the Error "No valid clips found for cut (duration needed)" is still an issue for lenghty PMVS & MUSIC.

Suggestion: The Audio or Video tab analyzes every lenghts of your Imported Videos so it knows what Timeline cuts it needs
= NO MORE "No valid clips found for cut (duration needed)" Errors.

Thank you, gooner689, for your suggestion, but I don't quite understand it. That error occurs when there are not enough processed videos in a segment of the Timeline to fill that segment. Until you try to create the timeline, you can't know if there will be enough videos. It will depend on the timeline you are trying to create. For example, if you set the Explicitness value to “Kissing” in a 1-minute segment, and if the sum of all scenes detected as ‘kissing’ in the processed videos totals 30 seconds, you will get the error “No valid clips found for cut X (duration needed)”. I don't know how you can predict that this will happen in the Videos or Audio tabs.

[clayk0]

this looks really awesome. The possibilities for the future with this seems really bright.
thank you for your hard work, already playing around with it.

questions: are there any limits with the procession and the created clips for a project?
I am thinking of starting to process more and more of my stuff and create a library that i can use for other projects without processing again.
I didn't try yet, but I can just copy the clip/frames stuff in a new project, right?

[MrLobo]

questions: are there any limits with the procession and the created clips for a project?

In principle, there are no strict limits on the number of clips you can have in a single project. The information extracted from videos is stored in a ChromaDB database. If you add an enormous amount of clips, it might take longer to perform semantic searches and consume more hardware resources when creating the timeline. However, ChromaDB is designed to handle a large number of entries, so you would need to add a vast quantity of clips for any noticeable degradation in performance.

I am thinking of starting to process more and more of my stuff and create a library that i can use for other projects without processing again.
I didn't try yet, but I can just copy the clip/frames stuff in a new project, right?

Currently, your project consists of two main components: the your_project.pmvd file and the your_project folder. You should avoid altering these files manually.
To reutilize videos processed in previous projects, you only need to load the existing project and then, from the "Videos" tab, add another folder for processing. This will incorporate the new videos with those already processed. Afterward, you can modify the audio and timeline as needed.
If you want to maintain the original audio and timeline settings while making changes, you can use the "Save As..." option from the menu when loading a project. This creates a copy of the project with a different name, allowing you to work on the new version without affecting the original one.

In the future, I plan to decouple the library of already processed videos entirely from the concept of a project. There will be a new view dedicated solely to video processing, where you can store your processed videos in a personal library. From there, you can import this library into any project and select which videos to use. This approach should make the process more comfortable and intuitive.
But I won't have this implemented in the short term; I need to change quite a few things for this. However, once the library with video information is decoupled, it will give me the opportunity to use it for a new function that I am planning. It's an AI mistress that generates tease videos using your videos, exploiting your fetishes and making reference to what happens in the videos. I already have some proof of concept, and the truth is that it generates quite interesting videos

[neo76]

Hello,
So I am trying it with only one video.
I did a little prompt and it divided the clip in 6 clips.
Whe I try to generate the timeline I got this error :

No frames found with acceptable semantic similarity

[17:56:46] Total segments: 6
[17:56:46] Segment 1: 43 cuts
[17:56:46] Segment 2: 25 cuts
[17:56:46] Segment 3: 23 cuts
[17:56:46] Segment 4: 14 cuts
[17:56:46] Segment 5: 14 cuts
[17:56:46] Segment 6: 8 cuts
[17:56:46]
[17:56:46] Processing segment 1: woman teases
[17:56:46] [WARNING] No frames found with acceptable semantic similarity. Using fallback frame (best distance: 0.630, threshold: 0.55)
[17:56:46] Cut 1/43: Distance: 0.630 File: 18OnlyGirls.12.07.08.Klara.Searing.Anal.mp4 - 00:40:907 (duration: 2.7s)
[17:56:46] Error generating timeline: No clips available for segment 1

Suggestions:
- Increase Explicitness or Nudity tolerance
- Enable 'Ignore filters when no valid clips'

[MrLobo]

So I am trying it with only one video.

Hi neo76,
you need to process more than one video because using only a single video will always result in an error message. When generating the timeline, the algorithm aims for variety by never placing two consecutive clips from the same video. This means that if you process just one video, all the clips come from it, causing issues when trying to create cuts beyond the first one since there are no additional source videos to draw clips from.

[neo76]

Thanks,
I am doing it with 7 clips and it works.
But, it doesn't really get what I want in the clip.
I mean, first scene is bj, to I put "oral sex" 1 woman but I still get hardcore in it.
Does the description in each clip count as prompt ?

Also, I don't seems to get as many frame as you, in a 22min video I got 19 frame with movement treshold @ 0.2, same with 0.50.

Even with max interval at 240sec it extract a 536sec frame somehow

[MrLobo]

Also, I don't seems to get as many frame as you, in a 22min video I got 19 frame with movement treshold @ 0.2, same with 0.50.

Even with max interval at 240sec it extract a 536sec frame somehow

I'll explain how video processing works so you can understand the screenshot log. It's quite detailed, but I don't know how to explain it more briefly.

Boring technical stuff about the tool:

Spoiler: show

First, the video is analyzed using ffmpeg's scene detection filter, which divides it into scenes. Then, these scenes are iterated over, and this is where the parameters you defined in the GUI under "Frame Extraction Parameters" come into play. The frames of each scene are then analyzed according to the specified "Target FPS." The similarity between one frame and the next is assessed, and depending on the "Movement Threshold," it's determined whether there is enough difference for a frame to be considered for analysis by the AI model. Even if the threshold is met, it's also checked whether the frame being analyzed is at least "Min Interval" seconds apart from the previously analyzed one. If the threshold isn't met but the last analyzed frame is more than "Max Interval" seconds old, the frame will still be analyzed.



In the screenshot you provided, the video was divided into 19 scenes, but this doesn't mean that 19 frames were analyzed; instead, 129 frames were analyzed. Out of these, 110 were saved to the database for use in the Timeline (19 were discarded, which happens when the AI cannot analyze them, such as when they are blurry or black screens). If you open your project folder and go to the 'frames' subfolder, you will see mosaics of your videos; each capture in the mosaic is what the AI has analyzed.

The AI provides a brief description of what's happening in the frame and marks an option for each of three characteristics:

Explicitness:
0- Normal pose, no sexual context
1-Suggestive pose
2-Groping body (not genitals)
3-Kissing
4- Touching genitals
5- Oral sex
6- Penetration (vaginal/anal)

Nudity:
0-Fully clothed
1-Revealing clothes (tight or short clothing, not underwear)
2-Underwear or swimwear visible (panties, lingerie, bikini)
3-Partial nudity (no genitals visible)
4- Full nudity (genitals visible)

People:
1 woman
1 man
2 women
2 men
1 man + 1 woman
3+ women
3+ men
3+ people of mixed genders
non-human or unclear

When editing a section in the timeline, you set the values for these categories and add a description. First, it searches for frames/clips that match these three characteristics; then, from the results, it performs a semantic search based on your description to select the best matches. After retrieving these clips, it filters out those that are semantically far from your description.

But, it doesn't really get what I want in the clip.
I mean, first scene is bj, to I put "oral sex" 1 woman but I still get hardcore in it.
Does the description in each clip count as prompt ?

If you set Explicitness to "Oral Sex" and People to "1 Woman," it's unlikely to return any results unless there is a woman performing oral sex on herself in your videos. For Oral Sex, it makes more sense to select multiple people under the 'People' category.

The generation of this Timeline should have failed, showing an error that no clips are available for that segment. If you didn't see this error, it's due to how you configured the filters below.



Explicitness and Nudity Tolerance allow flexibility in these categories. For example, if you set Explicitness Tolerance to 1, you're telling the system to look for clips one category above or below your specified value in all segments. Following your example, if you defined "5- Oral Sex" with a Tolerance of 1, that segment can also include clips from categories "4- Touching genitals" and "6- Penetration (vaginal/anal)". These tolerances are designed to help generate the video even when there aren't enough matching clips, albeit at the cost of strict adherence to the segment's requirements. You also have the checkbox "Ignore filters when no valid clips," which further flexibilizes segments by ignoring Explicitness, Nudity, and People if a search doesn't return results, performing only a semantic search based on your description. Lastly, the checkbox "Use fallback frame on no match" ensures that clips not matching your semantic description are not discarded (I recommend keeping this active). In essence, these options provide the flexibility needed to complete your timeline and generate the video even when strict segment criteria cannot be met, though this comes with the trade-off that the resulting video may deviate from what was originally defined in your segments.

I tried to explain all these details in the question marks for each parameter. If you hover your cursor over them, a pop-up will appear with a brief explanation.

In summary, to ensure a section contains only Oral Sex, configure the 'People' field correctly and set Explicitness Tolerance to 0. And make sure that “Ignore filter invalid clips” is unchecked.

Even with proper tolerances defined, you might notice some clips that don't match the segment requirements. This is due to the AI model miscategorizing frames. I've tested many models, and in the guide, I recommend the one that gave me the best results. However, I have not found any public models specifically trained for adult content, so they sometimes have difficulty understanding what is happening in a frame.

I hope this clarifies things rather than making them more complicated.

[neo76]

Also, I don't seems to get as many frame as you, in a 22min video I got 19 frame with movement treshold @ 0.2, same with 0.50.

Even with max interval at 240sec it extract a 536sec frame somehow

I'll explain how video processing works so you can understand the screenshot log. It's quite detailed, but I don't know how to explain it more briefly.

Boring technical stuff about the tool:

Spoiler: show

First, the video is analyzed using ffmpeg's scene detection filter, which divides it into scenes. Then, these scenes are iterated over, and this is where the parameters you defined in the GUI under "Frame Extraction Parameters" come into play. The frames of each scene are then analyzed according to the specified "Target FPS." The similarity between one frame and the next is assessed, and depending on the "Movement Threshold," it's determined whether there is enough difference for a frame to be considered for analysis by the AI model. Even if the threshold is met, it's also checked whether the frame being analyzed is at least "Min Interval" seconds apart from the previously analyzed one. If the threshold isn't met but the last analyzed frame is more than "Max Interval" seconds old, the frame will still be analyzed.



In the screenshot you provided, the video was divided into 19 scenes, but this doesn't mean that 19 frames were analyzed; instead, 129 frames were analyzed. Out of these, 110 were saved to the database for use in the Timeline (19 were discarded, which happens when the AI cannot analyze them, such as when they are blurry or black screens). If you open your project folder and go to the 'frames' subfolder, you will see mosaics of your videos; each capture in the mosaic is what the AI has analyzed.

The AI provides a brief description of what's happening in the frame and marks an option for each of three characteristics:

Explicitness:
0- Normal pose, no sexual context
1-Suggestive pose
2-Groping body (not genitals)
3-Kissing
4- Touching genitals
5- Oral sex
6- Penetration (vaginal/anal)

Nudity:
0-Fully clothed
1-Revealing clothes (tight or short clothing, not underwear)
2-Underwear or swimwear visible (panties, lingerie, bikini)
3-Partial nudity (no genitals visible)
4- Full nudity (genitals visible)

People:
1 woman
1 man
2 women
2 men
1 man + 1 woman
3+ women
3+ men
3+ people of mixed genders
non-human or unclear

When editing a section in the timeline, you set the values for these categories and add a description. First, it searches for frames/clips that match these three characteristics; then, from the results, it performs a semantic search based on your description to select the best matches. After retrieving these clips, it filters out those that are semantically far from your description.

But, it doesn't really get what I want in the clip.
I mean, first scene is bj, to I put "oral sex" 1 woman but I still get hardcore in it.
Does the description in each clip count as prompt ?

If you set Explicitness to "Oral Sex" and People to "1 Woman," it's unlikely to return any results unless there is a woman performing oral sex on herself in your videos. For Oral Sex, it makes more sense to select multiple people under the 'People' category.

The generation of this Timeline should have failed, showing an error that no clips are available for that segment. If you didn't see this error, it's due to how you configured the filters below.



Explicitness and Nudity Tolerance allow flexibility in these categories. For example, if you set Explicitness Tolerance to 1, you're telling the system to look for clips one category above or below your specified value in all segments. Following your example, if you defined "5- Oral Sex" with a Tolerance of 1, that segment can also include clips from categories "4- Touching genitals" and "6- Penetration (vaginal/anal)". These tolerances are designed to help generate the video even when there aren't enough matching clips, albeit at the cost of strict adherence to the segment's requirements. You also have the checkbox "Ignore filters when no valid clips," which further flexibilizes segments by ignoring Explicitness, Nudity, and People if a search doesn't return results, performing only a semantic search based on your description. Lastly, the checkbox "Use fallback frame on no match" ensures that clips not matching your semantic description are not discarded (I recommend keeping this active). In essence, these options provide the flexibility needed to complete your timeline and generate the video even when strict segment criteria cannot be met, though this comes with the trade-off that the resulting video may deviate from what was originally defined in your segments.

I tried to explain all these details in the question marks for each parameter. If you hover your cursor over them, a pop-up will appear with a brief explanation.

In summary, to ensure a section contains only Oral Sex, configure the 'People' field correctly and set Explicitness Tolerance to 0. And make sure that “Ignore filter invalid clips” is unchecked.

Even with proper tolerances defined, you might notice some clips that don't match the segment requirements. This is due to the AI model miscategorizing frames. I've tested many models, and in the guide, I recommend the one that gave me the best results. However, I have not found any public models specifically trained for adult content, so they sometimes have difficulty understanding what is happening in a frame.

I hope this clarifies things rather than making them more complicated.

I was using it all wrong !
Makes much more sense.
Thanks for taking the time to do such a good explanation.

I this possible to make a tolerance bar for each segment instead of the whole pmv ?
I got a teasing and then a bj segment
so I set the teasing one as 1-Suggestive pose and 1-Revealing clothes with 1 tolerance for both and it works nicely.
But then in the bj segment , since I put "oral sex" with 1 tolerance I have touching and sex in it too.

[kinkyhorizons]

Edit - I made some progress with it, I only seem to get the Filename or extension is too long when I'm trying to use longer audio tracks, did several around the 5 minute length with no issues at all, both that I tried that were over 10 minutes produced that same error.

I think I've managed to fix it. Now, for long videos with lots of cuts, the video is exported in multiple batches and then concatenated into a single video at the end, so that error shouldn't happen again.

Great, I've just downloaded the new version, will update to it and give it a try.

Again, fabulous work with this, it works really well and little niggles like this or the others people have come across aways to be expected once you release anything to a wider audience that might be using it in different way. None of them detract from cool a program you've pulled together.

[neo76]

So, I have been playing with it for some time, pretty awesome app.
I have a couple of suggestions if you can add it :

- tolerance and nudity flexibility for each segment instead of the whole clip
- An option where you can see how much cut you have for each keyword (ie: 76 cut - blowjobs, 150 cuts - non nudes, 206 cuts - sex man on woman, etc)

[MrLobo]

- tolerance and nudity flexibility for each segment instead of the whole clip

The tolerance design isn’t well thought out, it was a quick fix to allow video generation when there weren’t enough clips for certain segments, especially when Explicitness levels varied greatly across categories. I think it’s best to remove these tolerance sliders entirely and turn the Explicitness, People, and Nudity selectors into multi‑select options so you can pick multiple categories per segment. This will increase the number of videos that meet each condition and make the concept of tolerances unnecessary.
I hope to release a new version this week with this change and another feature that I am finishing testing.

- An option where you can see how much cut you have for each keyword (ie: 76 cut - blowjobs, 150 cuts - non nudes, 206 cuts - sex man on woman, etc)

This is not something I am considering at the moment; perhaps in the future. The next thing I will work on is completely breaking down the concept of video library and project. This way, there will be a new view where videos are processed (there will be greater control over already processed videos and it will be possible to view/modify analyzed clips), and the videos tab will simply import this library and select which ones you want for that project.

I believe this will allow more flexibility, as this library with video information and scenes can be used for other purposes and could even be shared (along with the analyzed videos) so that another user can use them without having to reprocess the scenes, which is the most tedious process and the only one that really requires a powerful GPU.

[neo76]

- tolerance and nudity flexibility for each segment instead of the whole clip

The tolerance design isn’t well thought out, it was a quick fix to allow video generation when there weren’t enough clips for certain segments, especially when Explicitness levels varied greatly across categories. I think it’s best to remove these tolerance sliders entirely and turn the Explicitness, People, and Nudity selectors into multi‑select options so you can pick multiple categories per segment. This will increase the number of videos that meet each condition and make the concept of tolerances unnecessary.
I hope to release a new version this week with this change and another feature that I am finishing testing.

This will help a lot, I am still struggling with the semantic search, can't get it to work properly, so I rely much more on categories.

This is not something I am considering at the moment; perhaps in the future. The next thing I will work on is completely breaking down the concept of video library and project. This way, there will be a new view where videos are processed (there will be greater control over already processed videos and it will be possible to view/modify analyzed clips), and the videos tab will simply import this library and select which ones you want for that project.

I believe this will allow more flexibility, as this library with video information and scenes can be used for other purposes and could even be shared (along with the analyzed videos) so that another user can use them without having to reprocess the scenes, which is the most tedious process and the only one that really requires a powerful GPU..

Thanks for all you are doing, excited to see where this is going.

[MrLobo]

Hi

I have uploaded a new version of PMV Director with two changes:

1- I removed the tolerances from the Timeline tab and, instead, it is now possible to select various categories of explicitness, nudity, and people composition per segment.



2- I added a new feature that allows for a 3-column layout with different vertical videos. To do this, there is a new option in the segments of the Timeline tab that lets you select the orientation of the videos to use:



- Any: Use videos of any orientation
- Vertical (Normal): 3-column layout with different videos
- Vertical (Mirror): 3-column layout, same video on left/right (mirrored)
- Landscape: Only use landscape videos

There is a new configuration variable in config.toml, under the timeline section, called vertical_cut_threshold_seconds. This decides how changes for these vertical videos will be made. If the cut is less than vertical_cut_threshold_seconds, only one of the 3 vertical videos will change; if it is greater, all three will change at once.

Here are two demo videos where I have used this new feature.

• Lina Belfiore compilation: https://www.eporner.com/video-dkg52NS07 ... mpilation/
  Backup: https://streamtape.com/v/K4Gkkbdz3bC0R9 ... o_Lina.mp4

• Ellie Leen compilation: https://www.eporner.com/video-UrSuzF3Z6 ... mpilation/
  Backup: https://streamtape.com/v/YDxLrQMrQ3C3YD ... _Ellie.mp4

I have updated the binary links in the first post to reflect the latest version.

[MrLobo]

Hi again

I am pleased to announce a new version of PMVDirector!!!

And this brings a major change: Projects and video libraries are now separate. A library holds your analyzed video data and can be shared across multiple projects. A project simply references a library instead of embedding the data.

Now, on one hand, we have the libraries, which are .vlib files containing the info about the videos we’ve already analyzed. And on the other hand, we have the projects, .pmvd files, which use the videos from the libraries to create a PMV.

These libraries can be used as many times as you like, and you can view and edit the videos they contain. The idea is to have a library containing information about your videos, and then, whenever you want to create a PMV, select which videos from your library you want to use.

From the new “Manage Videos” view, you can manage the videos in a library and even rearrange their paths if the videos have been moved to another folder.


By clicking the magnifying glass icon, you can view the processed scenes, including the description and categorization generated by the AI (which now allows you to test other AI models and easily see how well or poorly they perform). And from that view, you can also make manual adjustments, in case the AI made a mistake and you want to correct it.



Now, when we create a new project, instead of processing the videos, we simply select a library (vlib file) and choose which videos we want to use.

I've also added a button called “Preview Matching Clips” in the Timeline tab, within the segment settings.



This allows you to see which clips meet your segment's criteria. They're sorted in descending order, starting with the ones that best match your description.



I’ve updated the demo video with these changes: https://streamtape.com/v/rrPAjvY8erhx8J/demoV2.mp4

The video doesn’t explain all the options; when I have time, I want to create a more detailed tutorial that I’ll post in the first post.

I’ve updated the PMVDirector download links in the first post with this latest version. Keep in mind that older projects (.pmvd files) aren’t compatible with this new version.

Now that we have this library with the videos already analyzed, I plan to implement new modes to use it beyond just doing PMVs. I’m working on a “mistress” mode that uses this information from your videos to tease you.

Any suggestions are welcome