SLR Drive Vision and Future

lolitovtb Please elaborate on this... is there a guide somewhere?

Congratulations to the SLR team, I uploaded my videos and they worked perfectly. I hope all of this will be in production soon.

My opinion:

• I only use AI PT, I wish the processed videos could be downloaded, $9.99/video is a price I would gladly pay (I don't know anything about video editing, but I have enough local storage)
• I don't like recurring payments. I would prefer a lifetime payment for storage.

Thank you very much for making this technology available to us.

adinihsan max. 15m vid/720p for free

adinihsan thatoculusdude Earlier this year, before I even heard about SLR drive, I tried both RunwayML (paid version) and Da Vinci Resolve Magic Mask to create passthrough videos myself. While it's possible to do so with these apps, it's very impractical for any large scale conversion.
For RunwayML you need to downscale the video, upload it, create the green screen, download the green screen / masked video, upscale the mask back to the original resolution and overlay it onto the original video (using ffmpeg). Basically, it's too time consuming to do all of these steps.
For Da Vinci Resolve, the AI Magic Mask tracking can be finnicky with VR and too much manual input is required to make necessary corrections. Also very time consuming as well.
Realistically, for a large scale conversion I'm guessing you'd have to use AI models like Meta SAM2 (Segment Anything Model) in combination with YOLO model (You Only Look Once) for object detection but I never explored that fully. It would require a lot of coding and VR-specific AI model training to create a pipeline for large scale conversion with these models. I'm guessing nobody at the moment is willing to code such a thing for free.
Not to mention, if you don't have enough GPU processing for your needs, you probably need to spin up some beefy AWS EC2s with significant GPU / AI capable processing power - this can be very $$$.
So for now, we pay SLR.

GenericUser3 yeah. Nothing comes close to the convenience of SLR Drive. The only thing I'll keep doing on my own is VR deep fakes for personal use. I have a my own custom software and workflows in RunPod (expensive per single 8k vids, but worth it) and for the time being I haven't found anything superior. But PT and scripts? SLR seems ideal

doublevr THANK YOU. I built my toolset with Python and a lot of clever ways to refocus > dewarp > face swap > rewarp > paste back VR frames, and currently it takes me a few hours of full usage of 6-8 L40 GPUs for 40-50 minute 8K vids. The results are AMAZING, but the cost is a bit high. If you guys come up with something close to my solution and cheaper, I'd love you guys forever

Thank you SLR for this feature.. I uploaded and enabled PT, works fantastic.

would it be possible to introduce an archive option, which moves your videos to slower storage (like AWS glacier). Which should be cheaper and when you want to watch it, then you have the option to move it from archive to online. Maybe this can reduce the subscription costs..

doublevr No I dont mean replacing broken ones, Im just trying to clarify the note that pulled up right before I deleted the videos with credits applied to. It said it would take 24 hours for them to return to the account. What did that mean? it says

This action is permanent. Any scheduled AI requests will be canceled, and the tokens held will be refunded within 24 hours.

Should they return in 24 hours, or a month? I guess im not following

doublevr
Is this just for the "free" tier of Drive? Or are you saying that customers who pay for Drive subscription/credits have no refund recourse if SLR delivers a broken product?

doublevr
That sounds like a challenging decision.

On one hand you'd get a rash of customer complaints if the credits they paid for did not result in an acceptable product. But that would be gameable--once you've watched an AI'ed Drive scene, just request a refund with a fake complaint. Is SLR going to commit staff to double-check if the refund is justifiable? Or will SLR just let it ride and end up with increased AI processing costs because the abusive user runs through 60 AI requests in a month?

On the other hand, unlimited AI requests will clearly result in huge AI processing costs--which are likely to be passed to the customers. I would imagine that this would make the AI queue long AF and increase time from request to delivery.

--
How about allowing for automatic refund requests but flag abusive users and remove them from the Drive program?

Or maybe allow a limited number of re-requests for each credit spent on a scene? ForEx: I spend a credit on scene Alpha and the AI PT is shit. I can submit an AI request for Alpha to run back through the PT AI a second time for free.