Meshy to Blender workflow

Blender tutorial

My workflow to create models in Meshy and bring them into Blender.

What is Meshy?

Meshy is an AI app designed to create 3d models, texture them and rig them to animate them if needed.

It is mainly used to:

  • Create models for 3d printing
  • Create assets or characters for games

But it can also be used to create animation and (in my case) illustrations.

There are several of these apps around, which work the same way and have very similar interfaces like Tripo3D or Hi3d.

They generally have quite agressive marketing practices (spoofing the other apps names, retargeting, user-made promo videos). 

And you can generally test them for free with limited features.

Why use an AI 3d model generator?

A 3d model generator like Meshy will enable you to create complex models in a few minutes and texture them in a single process. 

For someone that is rather limited in his modelling skills (like me), it is very interesting.

A beholder I modeled in C4D being textured in 3DCoat
Meshy version of the beholder, done in minutes
  • Meshy also enables you to save money by creating models that would cost in the hundreds on Turbosquid. 
  • And there are things that have no models available anywhere. If you can provide an image, Meshy can create them.
  • It is far easier to import a Meshy model in Blender than to use a DAZ one, for instance (my alternative for characters). 

Advantages and limits in 3D creation

Advantages of Meshy

Limitations

  • Very good level of detail, starting with Meshy 6 and especially 7
  • Very good at creating a model from a single image and to "guess" what is at the rear
  • Correct knowledge of anatomy for common species. A horse looks like a horse.
  • Good for modelling details such as damaged garment, straps
  • Applies correct texturing to models. Recognizes metal in textures
  • Extremely detailed models are not easy to create and may have crappy textures.
  • Not very good for hard surface modelling
  • No design at scale
  • Animals or creatures modelled from a front view do not have correct proportions. 
    I call it the 'short snout syndrome'.
  • Some odd hallucinations on anatomy or technical details: does not know how a bow works.
  • Tends to have a "cartoony" look when creating objects from illustrations
  • Does not generate textures correctly from subjects seen from the front (more on this later)

 

The "social" part

Meshy is a subscription app that works with credits. You get a fixed number of credits each month, depending on your subscription, and spend them to create models. There are many features that aim at making it a community of users.

Once you have your paid account, you you can greatly extend your possibilities.

For instance you can: 

  • Browse models made by others in the community and download them for free. 
  • Share your own models in the community and earn some credits
  • Follow other users to find what they are doing.
    There are some very talented artists such as Bill Stelling, and creators specialized in specific topics like PICKTURA.
  • Create lists of models to easily find them afterwards and follow lists created by others. This is especially useful as the models search engine is not very good.

This is an important part of the tool that you should consider when subscribing.

 

Models form the community: friendly goblins from PICKTURA

The process of creating a 3d asset

Before starting any creation process, check the community for models that may suit your needs. This could save you time and credits.

If you have collections, check them for the models you stored.

Here is my workflow to create models.

1. Find inspiration

I generally head to Pinterest to find images that may be used to create models. Ideally, the subject should be fully visible and seen from three quarters front. But that may be changed.

Here is an image I have used. It is the cover of a book, an illustration made by Angus Mc Bride.

My aim is to create a model from the mounted horseman.

2. Adjust the source image 

I need to adjust the source image to make it suitable for model creation. 

To do so, I generally use Gemini image. Meshy has an integrated version of Nano Banana, but it costs credits and does not allow image adjustments.

My first aim is to keep only the horseman and remove the other guys.

Prompt:Isolate the mounted warrior in blue armor on a white background, without the other two. 

Result: (it kept one of the footmen)

Prompt: Remove the footman

Result: Gemini re-created the missing part. It also added a shadow, but this is not a big deal.

We could work with this, but it would create a model that looks a bit like a cartoon. 

The next Step generally works wonders.

Prompt: Make this illustration photorealistic and remove the shadows on the ground. Keep it on a white background

Result: this will give a far better 3d model.

It is now time to head for Meshy. We will use this picture as a template for our model.

3. Create the model

In Meshy, input your picture for the model creation. 

For such a character, I chose the following parameters: 

  • Model: Meshy 7 (the most detailed so far)
  • Resolution: Ultra 4K (not 100% needed)
  • No texture (important)
  • No specific pose
  • Image enhancement

After a few minutes, Meshy creates a blank character. Once you are at this step, check your model carefully. If it has errors, you may retry to generate it for free. This is why you should not create the model with its texture in one step.

Once you are satisfied with your model, you may continue to texturing.

Use the same image as a texture base.

 

You can notice that there is a slight error on the model: the armories on the shield are blurry. This happens when using a reference image showing your model too much from the front. I tried to re-generate the texture, but got the same result. It did not bother me much as I was going to use the character from the back, but I could have fixed it in Blender.

The model is very detailed and has 1.9 million triangles. This may not be suitable for every use, so Meshy proposes a retopology option, which is quite efficient, but tends to remove some details.You can choose to switch to quads.

Our model went from 1.9 million triangles to 113 000 quads and has a better topology. 

We can now take it to Blender.

Importing into Blender

There are two ways to import a model into Blender.

  • Save it as an fbx and drag and drop it into Blender
  • Use the Meshy DCC Bridge

The latter option does not yet work with 4K models.

Installing the bridge is quite straightforward. You will find instructions in the documentation.

Please note that you will have to enable the glTF 2.0 format addon in Blender for the bridge to work!

Our knight is now in Blender and ready to fight.

Other models

I share this model and many others on my profile in Meshy. Come and take a look!