i can't find the perfect settings about depthflow HOT 12 OPEN

I've got AMAZINg results with some simple filtering with MaxFilter kernel (to "fatten the edges") and GaussianBlur for smooth edge transitions + the upscaling (not noticeable here as output is 1080x1920)

still old / current animation; solving some more code logistics and thinking where to place stuff~ !

from depthflow.

After analyzing some other user's video on Discord and trying to match their intended effect, I've discovered one new useful parameter to center the displacement on off-center objects, perfect and needed for your image!

I'm documenting on the code what all parameters does and implementing this new thing (+depth focal point)

I will commit those in an hour or two if things go right, but until then what camera animation are you looking for in this image?

eg. background or foreground stationary? natural "camera shake"? any form of dolly zoom effect?

from depthflow.

I want to create same like the "leiapix.ai".

this is the video created by leiapix but the problem is with the quality and those annoying watermarks.

this is the depthmap of the picture. Depthmap is not the problem. The main problem is only the proper settings to create like this and its hard to find the proper settings.

If you can tell me about the upscaling then it can also help me.

from depthflow.

Give me an hour or two, I think I can work myself through a 'reverse engineer' of the shaping functions on parameters this animation uses!

from depthflow.

I'm trying ways to fix the edge artifacts that happen because depth anything's estimations are a bit too "thin" and abrupt. It's getting there

from depthflow.

yeah please and what about upscaling. How it works?

i remember the depthy website but the video process doesn't works anymore on that.

from depthflow.

upscaling would work by configuring a class of mine and running image = upscaler.upscale(image), and optionally a output=Path to cache the result, else it creates temporary files for any operations

upscaling the input won't make the depth estimation better as most estimators downscale the image (also part of the jagged edges problem), but I've got nice results in upscaling a slightly blurred depth estimation itself and using it

~

I'll give some love and hook up the upscaler classes as well for manual use as it's an easy-ish thing to do (+download the executables from github), I'm facing some really interesting issues regarding cross-packages type checking

I'm diverging from your request but this upscale step will make the final result much better, it's worth it

from depthflow.

when I say upscaler I mean realersgan x4 plus anime model- !

from depthflow.

Just got upscalers back working, will improve, review and test parts of the code, eta a few hours

Not sure how extensible the CLI should be for them, maybe just the ratio, and if it's 1, then don't upscale (default), but stretch/downsample whenever needed, the visual improvements are big specially for low res AI images!

from depthflow.

I've encountered a similar problem, and even the generation of leiapix.ai doesn't satisfy me:

Can we split the image into different depth groups, then use image ai to fill in the missing content at that depth group and generate the video after that?

from depthflow.

I've got AMAZINg results with some simple filtering with MaxFilter kernel (to "fatten the edges") and GaussianBlur for smooth edge transitions + the upscaling (not noticeable here as output is 1080x1920)

still old / current animation; solving some more code logistics and thinking where to place stuff~ !

depthflow-12.mp4

what are settings bro. Please share the settings. This is what i actually wanted

from depthflow.

I've encountered a similar problem, and even the generation of leiapix.ai doesn't satisfy me:

Can we split the image into different depth groups, then use image ai to fill in the missing content at that depth group and generate the video after that?

i think it only works with one depth group.

from depthflow.