yeah, things change very fast in the world of AI.

I answered in detail but KZbin deleted it. It was step by step comment I posed here. Pity you would have know how to do it.

yes, i believe so. You may have to google on that one though. Note that it will most likely be done frame by frame. Maybe huggingface has something or somebody might have to put a python notebook to do just that on google colab. I know some pple do video conversions like Philip Heggie (he's on youtube so you can ask him his process).

94 lpi??? Are you sure??? You can do good stuff with inkjet if you are under 50 lpi. For animation, it's difficult if the images vary rapidly, like for instance, a batter swinging at a bat. You ain't gonna see the bat moving basically. If it's slow moving, should be ok. You can definitely try stuff at home, just use good quality paper and make sure you do a pitch test prior. I am really not an expert at doing lenticulars.

Thanks. In this day and age, it's probably easier to use AI to generate painterly images from photos. But, to try to answer your question, if you look at the input file github.com/ugocapeto/thepainter/blob/main/main/test/waterlilykiwi/thepainter_input.txt, you can see that the number of brushes used is 5, starting with a radius of 128 pixels, then 64, then 32, then 16, then 8. The lower the brush radius, the more detail you are gonna get. Also, the 4th number, here, 50.0, indicates how close the painting should be to the original photo, so is you put 20.0 instead of 50.0, you are gonna get more brushstrokes. What I used to do is create several output paintings at different level of resolutions and then use Gimp to combine the paintings. Hope that makes sense. Note that the code is available on github at github.com/ugocapeto/thepainter. Also, note that I haven't touched this stuff for a few years, so i don't remember all the details but i do try my best to explain what I recall :)

Things change very rapidly. You can try hugging face spaces that relate to stable diffusion: huggingface.co/spaces?sort=trending&search=stable+diffusion. You are gonna need an image-to-image and in the prompt, put "oil painting" or something like that.

I answered in detail but KZbin deleted it. It was step by step comment I posed here. Pity you would have know how to do it.

yeah, i don't know what's going on with those runtime errors. All the midas stuff on hugging face have runtime errors. Anybody knows why???

2 flip should work as long as the dpi is not too high. Higher lpi (like 75) is difficult to deal with. If you have more images to flip, you're likely to get ghosting. But a 4 image flip should be ok, i think. 8 is pushing it especially if the images are completely different. With 60 or lower lpi, you should be able to do animations with 8 images but that assumes the images are not too different from each other. If you do an animation of a bat swing in baseball, the bat movement will not be sharp. Not that I am not an expert in making flips but I have tried to do animations with 60 lpi, and it's quite difficult if not impossible with an inkjet printer.

yes, correct. The only rule to follow, If you use inkjet printer, is that you need to print so that the "stripes" are 90 degrees w/r to the print head carrier. in other words, You don't want the stripes going in the same direction as the print head carrier. This gives you better print for lenticular. At least, that's what I have always been told.

got it, many thanks.@@ugocapeto3d

Check www.dropbox.com/s/wsuelhwgxnj8a6n/ugosoft3d-11-x64.rar?dl=0 This archive contains SfM10 and MVS10, it will have the manuals in pdf form. 3dstereophoto.blogspot.com/2016/04/structure-from-motion-10-sfm10.html 3dstereophoto.blogspot.com/2016/04/multi-view-stereo-10-mvs10.html

Glad it could be of use. I had fun making the video as I love making "paintings" from photos.

Looks like Colab broke the notebook with it's update. I fixed the issue by reverting to a previous version of Torch.

Hi, don't get your hopes too high. The depth maps obtained will, in most cases, require hand-tuning. But Midas v3.1 is the most advanced ai tool to get depth maps from single images. As it gets updated, it gets better and better. But the updates don't come out too often.

@@ugocapeto3d I suspected as much, over the many past years things get close but not close enough. Hand tuning was a must with everything else I have seen previously (non AI). Zbrush does a wonderful job with 3d setup scenes in creating a bas relief almost perfect. I have seen midjourney images of depth models/bass reliefs that look amazing but I suspect only for viewing and not for real world application. Ty for your comment.

Thanks a lot!

Sorry, this was done at a time when annotations were still a thing on youtube. Anyway, this is old tech. I recommend using automatic AI tools now, like Midas V3.1. See this vid: kzbin.info/www/bejne/iGi7mKB5lL2opaM. If you want more ease of use, you can use LeiaPix converter although it is based on an earlier version of Midas: kzbin.info/www/bejne/oKLGlGuFlrtpeNE

Check the github repo: github.com/isl-org/MiDaS. At the end, there is a link to their latest paper.

@@ugocapeto3d well I went through said paper but I still struggled to actually understand how they got their loss function and such

Thanks a lot for your comment!

I recreated this using ffmpeg minterpolate. It's much slower and the results are more datamosh-y. But still cool.

You're welcome!

You can download obj with depthplayer.ugocapeto.com but you will lose all the colors.

@@ugocapeto3d yes, i already did that, but when i open the .obj model in blender, it appears with the "solid" render mode. For the colors i just have to put the base image as a material.

If I remember correctly, you can use your own depth maps if you use the implementation that's on google colab. I remember making a video about it.

This one: kzbin.info/www/bejne/q4PXoKVrepKdpMk at around 18:52, I use my own depth map.

You need to download the nogui archive. You must have downloaded the other one. Follow this tutorial by the SPM creator if you have difficulty (in particular, step 4): www.stereo.jpn.org/eng/stphmkr/makedm/index.html. Note that this is to get depth map from stereo pair. If you have a single image, I recommend Midas. See this video: kzbin.info/www/bejne/jXuxgKGCmb5sgck where i compare spm dmag5/dmag9b and Midas V3.1.