From text-to-image to a fully controllable 3D logo

The best thing about Stable Diffusion v2's depth2img feature is that it gave us another tool in our arsenal and it opened up our eyes what the so-called depth estimators can do. So over the past few days we've been playing around with these depth estimators and we managed to create a workflow that lets you create 3D logo animations very fast.

Here are the results.

The workflow described in this article lets you create fast logo prototypes that are not just simple 2D images but they are fully controllable 3D assets.

In this guide we walk you through this workflow so you can replicate the video above. We hope you'll leave with a ton of value.

We'll share all the details but before we do that let's go through...

The main idea

Alright. In order to have a fully controllable logo, we need a couple of things.

First we need a logo - that's easy. We can use any of text-to-image generators (Stable Diffusion / DALLE / Midjourney). Then we need a depth estimator that can tell the estimated 3D shape of our 2D file so we can tell our modelling software how our logo looks in 3D. We also need a background remover and a pixel to raster image converter - these might not be obvious right now, but later it'll become clear why we need them.

...And these all need to work beautifully in tandem.

Challenge accepted - let's get started!

Give me a logo

First of all, we need a logo. Fortunately that's easy to create.

We decided to use Midjourney v4 to test out its logo creation capabilities but you can use any text-to-image generator.

Many people wrote about how to generate cool looking logos with Midjourney so we're not going to reinvent the wheel. If you're interested about the dark secrets of logo creation or looking for some prompt inspiration, check out this page - it has many great examples.

We decided to go for a logo with an octopus and used the prompt flat logo design of octopus on a shield, flat 2d, sharper, company logo --v 4

We liked this one.

(Let's ignore the text for now, you can easily change that with tools like Photoshop or Figma).

Depth estimation

Depth estimation is the process during which an algorithm tries to estimate the 3D shape of our 2D picture.

This is still difficult and requires a lot of trial and error. And patience.

During this project we used several depth estimators and the conclusion is that you really need to try all of them during your projects because they perform differently in different scenarios. A few depth estimators we tried: MiDaS, DenseDepth, Dense Prediction Transformer.

For our logo creation MiDaS worked the best. We found that removing your background with an AI tool like remove.bg also improves the output. So we recommend using it before the depth estimation.

Alright. We've got the raw file, the logo with the removed background and the estimated depth. Fantastic. We're almost ready to open our 3D modelling software!

The last thing that we need to do before that is to convert our logo to a vector file.

To do this first we first applied some filters on the image in Figma to have a more-or-less black and white photo with the logo and the background clearly separated. You might need to apply these filters several times depending on your logo.

Finally we used Convertio to turn our png file to svg.

We have everything now to get started.

Awesome.

Turning 2D files to 3D

Alright, here comes the real fun.

Let's open up Blender and remove the basic cube.

We first need to add a plane (Shift+A -> Mesh -> Plane). Right now it has only two triangles, so let's select it and hit Tab to edit the mesh.

We can right click -> Subdivide to increase the resolution of the mesh.

Let's subdivide the plane 9 times so we had a 512 x 512 mesh. Then hit Tab again to go back to object modelling mode.

Once we have a fine mesh we can add the texture as well as the depth map.

Let's go to the Shading tab and add a Principled BSDF material with an image texture node. If you choose the logo with the removed background as the input of the image texture node you should see something like this:

The next step is to add the depth map to the plane. Let's go back to the Layout tab and add a Displace modifier to the plane. Create a new Texture.

Go to the Texture tab and add your depth map. In the Sampling section check in the Interpolation box and increase the size to a value that you like (in our case 10 resulted in a smooth mesh).

If you did everything the right way you should see something like this.

Fantastic. We're getting close.

We now need to remove all the parts of the plane that is not part of the logo and this is the reason why we need the SVG file we created earlier.

Once we have the SVG imported, we can extrude that and apply a boolean modifier on the logo and the extrusion.

Let's do this.

First we import the SVG file (File -> Import -> Scalable Vector Graphics) and make the displace modifier on the plane invisible.

Our SVG is pretty small and isn't at the right place. So let's change that.

We need to position and scale the SVG such that it covers the logo on the plane.

Once we're done with this let's extrude the curve (if you have more curves in the SVG select them and merge them with CMD+J).

You should see something like this.

Convert the curve into a mesh by clicking on Object -> Convert -> Mesh and reactivate the Displace modifier.

Finally, apply a Boolean modifier with the intersect option on the plane. Select the curve as object.

Hide the SVG mesh and you should have your logo in 3D.

As the last step we added a background and lights that fit the 3D logo.

And the rendered version:

We quite like the end result.

Of course it's not perfect but it is a nice workflow to create fully controllable 3D logos easily by mixing text-to-image generators and 3D modelling software.

Hope that you liked the outcome.

If this article delivered value to you subscribe to our newsletter to receive more workflows like this in your inbox every week.

‍

Gabor

Founder of Generative Nation