Create Better Quality Depth Maps From 2D Images With New AI Tools 

BTW MIDAS 3.1 LOOKS BETTER BY A LONGSHOT. PS FANBOYS COPING HARD!

Siikoako

Awesome, thanks so much for subscribing.

@@AIAnimationStudio you are welcome. But I have one question. I have just tried everything that you showed us here with AE and my camera doesn't work. Actually, orbit-around option doesn't work. thanks in advance.

@@tanyasubaBg . Hey, it should work as expected. Feel free to email via the aianimation.com website and I'll try and quickly help. It may be the 'constantly rasterise' button needs pressing or unpressing. I think (🤞) all the steps and button presses should be covered in the video. (Also the setup in this particular video needs the camera to use the pan tool not orbit)

@@AIAnimationStudio thank you

Hey@@tanyasubaBg , just looking back at this, I think it might be a simple case of you're using the incorrect camera tool. When pressing 'c' to cycle through the camera move tools, you'd want to use the 'pan' tool in this instance, not the 'orbit' tool, as the displacement map is connected to the position of the camera and not the orientation. Hope that helps.

Great idea. I've added the dystopian cityscape, plus another image to download for free via this link aianimation.com/tutorial-06-create-higher-quality-depth-maps-with-ai/

Yeah, that's on the list to do a video with, but wanted to take a non paid plugin approach. I'm interested to see if their 'accurate' and 'highest quality' settings compare.

Yeah, you can definitely do stuff like that. (Good old Andrew Kramer Tutorials... 🙌.. I might have to see if I can find his old tutorial to see exactly what you mean). I assume it'll be a similar approach to dropping the character into the 3D space in this video. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-t5Vq4ahmn74.html

@@AIAnimationStudio ive seen others do it. Basically imagine a cube in an empty room. Now you wanna create realism by having bursts of flames and smoke. In real life the flames & smoke would be infront and behind the cube. I think you would use levels / curve sliders on the depth map and this would cut out the foreground - mid - background areas. In this video in the photoshop AI filters, im sure its using this technique to generate the depth of field . Its applying a mask to the blur in foreground - mid - background.

@@AIAnimationStudio ah. Could it be gradient wipe. Great example here of the masking. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-VJJkCRHqyno.html Ffw to first layer chapter at 2:34

​@@armondtanz .. Cool!!.. I know exactly what you mean. Great idea for a future tutorial, it's going on the list. I've actually got it going on in an AI short film I'm 'slooowly' making at the mo, with robots in amongst jungle leaves. It's a really useful technique and great way to relatively easily create something that looks a lot more complex.👍

Thank you, glad you liked the tut. Unfortunately can't paint the depth map to fix the wonky parts, as it's more just a limitation of the effect. I've been trying out different approaches, i.e. with the more advanced (but free) Displacer Pro plugin for After Effects to see if better results can be obtained. Plus some other ideas, which I'll share if they show decent results. Plus you could make a 3D mesh of the displacement via Blender and bring that into After Effects (potentially via the new 3D file support in After Effects Beta, or use Element 3D) but that still has similar distortion and adds more steps to the workflow.

@@AIAnimationStudio okay, bummer. I thought you could since the depth map appears to be differing values of black and white, but I guess there’s more to it. Thanks for the quick response and keep up the good work!

@@AIAnimationStudio Curious what your results were with Displacer Pro. Any improvement? Or have you tried AE Depth Scanner? OH....and GREAT tutorial videos dude. Love your stuff!! ...and I'm not gonna be lazy. I'll try the Displacer Pro myself and see.

@@AIAnimationStudio As a follow-up (to my trial with Displacer Pro), I found the camera movements to be WAY too sensitive. The slightest mouse movement made the effect distort completely. So I wrapped each of the two Camera scripts in brackets and divided the total by 50 (or 100 for even smoother control). So, this is what changed: thisComp.layer("Camera 1").transform. Position[0]-2000 --> thisComp.layer("Camera 1").transform. Position[0]-2000 The 2000 is just my particular X positional value so will vary for anyone trying this themselves. Then repeated for vertical script. Works like a charm!

Nice, thanks for sharing that. @@rkgibbons. I got distracted by some other parts of production but will no doubt revisit depth maps again soon as it's such a useful approach. Have you tried out the 3D format import into the After Effects Beta, that adds another alternative approach for add depth to scenes, with a 3D mesh made via somethinglike Zoe Depth. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7u0FYVPQ5rc.htmlsi=KOL3yREamqGeAmKf&t=771

Considering Depth Scanner includes MiDaS, ZoeDepth, and now Depth Anything builds in their models, I find that notion kinda silly. What Depth Scanner does provide is a better post processing pipeline for editing and 3D conversion.

sometime

🤟

If installed locally, It really should not have taken any where near that amount of space on your C:: disk. I can only assume something else unrelated happened. Good luck finding the cause. (But you could use Midas via hugging face, or Colab without filling up space on your harddrive (bar the small file you download at the end). Also I'd now recommend checking out ZoeDepth which you'll find on Hugging Face as well.

same here! not sure what the error means TypeError: expected size to be one of int or Tuple[int] or Tuple[int, int] or Tuple[int, int, int], but got size with types [, ]

same here..

Does anyone know if there is a fix for this? are there any good alternative collabs that can be used instead?

Colab 3.1 version accepted my 3K input image and produced a fine quality output, without downsizing.