Use AI to Create Depth Maps | After Effects
HTML-код
- Опубликовано: 20 июл 2024
- You might have heard of depth maps... but have you used them in your projects? This episode takes a look at what depth maps are and a few cool things we can do with them- aaand best of all generate them easily with AI!
You can use this AI model to generate depth maps: huggingface.co/spaces/toshas/...
If you followed this video don't forget to tag me on Instagram @jumpyvertex, I'd love to see your creations!
👉 Follow me on socials at @jumpyvertex 👈
Please don't forget to smash the like button and subscribe, your support means a lot and makes more videos possible!
Credit in order of content appearance,
Photo by Vitaly Gorbachev from Pexels: www.pexels.com/photo/low-angl...
Photo by Eric Prouzet from Pexels: www.pexels.com/photo/miami-be...
Photo by Antonio Cuellar from Pexels: www.pexels.com/photo/building...
Video by Maksim Romashkin from Pexels: www.pexels.com/video/a-house-...
![Chk Chk Boom - Stray Kids ストレイキッズ 스트레이 키즈 [Music Bank] | KBS WORLD TV 240719](http://i.ytimg.com/vi/63QPwqCRGcw/mqdefault.jpg)
Gave a thumbs up right away for the avatar :D
Found your video while scrolling Reddit, found the RUclips profile in your profile, and found this video.
The video is so professionally made, so clear, so informative plus you manage to show us your sense of humour too through it!
Wish you loads of success and looking forward to seeing more and more such content. (Subscribed just looking at the thumbnails)
Thank you so much for your lovely comment and subscribing! So glad to hear you’re enjoying the videos! 😊
I think this is just a spot-on tutorial, brilliant job!
Good stuff, man. Thanks!
Wow!
First of all: congratulations for the tutorial. I really loved it and it can help me decide if I go pay a hundred dollars for a plugin to do depth maps because the trial version really doesn't helped me though.
And congratulations for being the best and more entertaining tutorial to watch haha
Loved how you animated your character and make it go alive!
Thanks so much! Glad it helped you out!
Brilliant- but did you drag EACH frame one at a time into Marigold?? YIKES... that's prohibitive unless it can do batches. Tho maybe we could do it in Photoshop Depth Blur for batch frames.
Thank you! For sure, wouldn’t recommend it for lengthy shots - it’s practically free tho. BUT there are other new paid plugins out there that do this much faster if you’re in a hurry.
Wow! Cool video! Thanks!
Glad you liked it!
Just use depth scanner from aescripts
Kinda pricey but pretty great
Have you used depth maps before?
can you make a lesson on this guide?
''We begin the process of creating a depth map using Artificial Intelligence for image creation from text, there are many options available at the moment, and personally, I prefer to use Bing Image Creator because it can interpret my intention of making deep 3D engravings very well.
The main components of a good prompt usually include:
Subject description: Detail what you want to appear in the image. Be specific about the objects, people, animals, or landscapes.
Artistic style: Indicates the artistic style you desire, such as watercolor, photography, pixel art, etc.
Composition and context: Describe the composition of the image, including lighting, framing, and ambiance.
The clearer and more detailed your prompt is, the better the AI's understanding will be, and the more accurate the results will be.
In our particular case, it is very important to mention that we expect a visual result similar to wood carving in high relief. Terms such as 3D rendering, thick details, frontal view, CNC Router, the shape of the final figure, let's say a round medallion, coin-stamping style, hyper-realistic, color depth, light orientation, and why not, mention a depth map in grayscale.
These are just some of the terms to mention, but one must be careful not to overdo the descriptions because overly long prompts can cause the final intention of a 3D image to be lost.
In some text-to-image AIs, you can write the subject within parentheses to emphasize that this is the most important part of our prompt.
Using image editing software
Once we have managed to generate the image that we like the most, it is important to convert it to BMP format in grayscale with 256 levels. You can use online image editing programs like PicWish to enhance the image quality. It is necessary to edit the image to have a solid black background around the medallion image to be engraved.
The next step is to use a depth estimation application like ZoeDepth or Marigold to generate a primary depth map. This initial depth map lacks details and when tested, it appears as a blurry figure with very few protrusions.
To add details and clarity to our depth map, it's necessary to blend our primary depth map with the original image. For this purpose, we can use programs like Photoshop, Corel's PhotoPaint, or the free alternative Gimp.
The process is the same in any of these programs; you need to create a new document with 600 DPI at least twice the size of the final file we expect to obtain.
We open the image created with the depth estimation app, convert it to a 600 DPI BMP, in grayscale with 256 levels, with an exact measurement in inches equal to the original file created in Ai from text to image. Most importantly, we use the 'Invert Colors' option, making black become white and vice versa. It's important to remember that in most cases, when creating a depth map engraving, white represents the highest points.
Once we finished doing this changes it is saved as BMP file.
The next step is to edit the original image created with the AI text-to-image to increase brightness and contrast. With this, we aim to create a high-contrast image as much as possible without losing details. The greater the difference between the lightest and darkest colors, the greater depth we will achieve in the engraving.
Now, we return to the blank new document and import the primary depth map created with the depth estimation app. We center the image and then import the original image created with the AI text-to-image on another layer, ensuring it covers the primary depth map image perfectly.
To avoid excessive noise in the background of the primary depth map, we create a third layer and place a solid black figure with a circular window the same size as the circular medallion. This covers the image with solid black but leaves the area to be engraved uncovered.
With the three layers in place, we select the layer of the image created by the AI text-to-image and apply transparency, reducing it from 100% to between 10% and 20%. By doing this, we can see the details of the original image through the primary depth map image.
At this point, we export this image as a 256-level grayscale BMP at 600 DPI.
The process can be considered complete at this stage, but the depth map can still be slightly improved using a depth blur filter. This procedure is somewhat controversial as achieving the best results isn't easy, but when applied correctly, good results are obtained. This step is optional.''
Upload each frame??? No way!
But, how to set up it offline, is that with Python? Make another tutorial! :D
Will do!
can we do this depth map creation locally like automatic1111? Because ive got this cool nvidia card
Can do! You’ll need to download the model from the same link and set it up for offline use
did you say to upload each frame !
hahahaha
Can't I use davinchi for the depth map ??
It can make it I think
Sounds cool, haven’t tried that yet but will check it out!