Reference images
Hello. Thank you for your work.
As I understand it, your model currently works only with FL2VA models.
This means it can’t work with reference images, which is a sad..
I think many people would agree that Ref2VA, thanks to the ability to add character references (body proportions, facial angles, emotions, and so on), does a great job of generating consistent frames with the character, something FL2VA copes with much worse.
But as I understand it, without FL2VA it’s not possible to generate sequential frames.
I would at least like to have the option to generate the initial frames using Ref2VA and then continue with FL2VA for the subsequent frames.
Ideally, I’ve already seen some models and workflows that allow you to combine both initial frames and reference images, although I’m not sure how stable or reliable this approach would be.
Right now, though, the ability to add additional reference images is something that is really missing.
I have tested the FL2VA with this method and it works as designed. People would be pushing for REF2VA as they want to keep creating long extended anime with it, which the Internet is already filled with.
I have a philosophy that content should be original and not re-hashed from copyright works. That is also part of the intended design.
I have a philosophy that content should be original and not re-hashed from copyright works.
I don't really understand what reworking/reprocessing artworks has to do with this.
And how the absence of REF2VA would help get rid of all of this, considering the variety of services and models.
And in what way FL2VA differs from REF2VA in this regard.
It's simply an additional way of maintaining control.
It never even crossed my mind to upload any copyrighted works somewhere.
So I really don't understand what you mean, but it's your call.
I have a philosophy that content should be original and not re-hashed from copyright works.
I don't really understand what reworking/reprocessing artworks has to do with this.
And how the absence of REF2VA would help get rid of all of this, considering the variety of services and models.
And in what way FL2VA differs from REF2VA in this regard.
It's simply an additional way of maintaining control.
It never even crossed my mind to upload any copyrighted works somewhere.
So I really don't understand what you mean, but it's your call.
Well at least you have ethics in regards to that. I applaud you in that respect. The 99th percentile is going to use it for Tiktok/YT anime slop, but I digress.
I have created a REF2VA version and applied it to this repo. It can also be pulled down through the main Github repo. I am not sure what the output is going to be like, but it will support up to 4 reference images to start.
Unfortunately, there’s nothing that can be done about it now, but I’m sure it will all sort itself out organically over time.
People are already tired and sick of this kind of content anyway.
As demand declines, all this crap will either die down or evolve into something else.
I don’t think it’s anything more than hype, like NFTs, crypto, and so on were back in the day.
But thanks for the REF2VA version, I appreciate it.
Well even though Vines pretty much died off, memes never did haha.
Time will only tell.