Presented by
Autodesk Flow Studio challenge: Can I design a free 3D model with just 300 credits with Wonder Tools?
Over the last couple of articles, we’ve explored together in detail how to use Flow Studio’s new Wonder tools - which are also available to non-paying customers.
That free tier is a great way to get introduced to everything Autodesk’s Flow Studio has to offer, but just how much can you really do with it? I mean, you can kick the tires and create a few sample images for sure, but can you produce something that you can take away and use in another software package, either to refine further, or insert into a video project perhaps?
That’s the goal I’ve set up for myself here: to see if, at the end of the day, you can achieve what you need even on the free tier.
A Brief Recap
Before we go any further, I think it would be sensible to offer you a quick recap on what I’ve discovered so far, should you be new to my Flow Studio experiments (you can read my past adventures here and here).
Put simply, the new Wonder tools are Generative AI: you use a text prompt to create an image, or either upload a photo, or simply use that text prompt again to create a 3D model, complete with texture. It’s also possible to edit and further refine something already generated.
In order to do this, you need to spend credits. A free account gives you 300 each month, and each step in the creation process will cost you 20 credits - meaning in this challenge, I have a maximum of 15 steps to complete my project.
Of course, should you choose to pay, your credit allowance will be much higher (along with additional benefits, such as higher resolution exports for one, and the ability to get the algorithm to work on more than one model at a time for another). Tiers range from $7 to $64 per month paid annually, and there’s even custom pricing for those needing even more bang for their buck.
But that’s all by the by: my goal is to see how far my 300 credits will take me.
Getting started
Start by heading to Autodesk’s Flow Studio’s webpage by clicking here. Then either log in or sign up for a free account.
You’ll be taken through the registration process, which only asks for an email address and password. No sneaky credit card request which can potentially lead to even sneakier unexpected enrolment to a paid tier at some point - that’s not how Autodesk roll.
Flow Studio’s home page graces you with featured templates and characters (I explored what you can do with those in this post), along with any recent project you might have already worked on. To the left is a sidebar, where the Wonder Tools can be accessed. If you can’t see that option, scroll down the section bearing the Home and My Projects buttons to reveal it.
The three tools I could potentially use for this experiment are ‘Text to Image’, ‘Image to 3D’ or ‘Text to 3D’. If you’ve got a photo of an object you’d like Flow Studio to transform into a 3D model, that’s likely one of the easiest commands you can give it.
But in this case, I don’t have a starting image: I’d like Flow Studio’s Wonder tools to create a futuristic flying car of me, something that’s different from other such concepts we’ve seen in movies, like Back to the Future or Blade Runner (as we certainly don’t want to run foul of movie studio lawyers - you know how they can be).
So let’s start with ‘Text to Image’ and see what these algorithms can come up with.
Text to Image
If this is the first time you’re launching any of those Wonder tools, you’ll be reminded that you need to own the rights of any image you might be uploading to this service: Autodesk want to stay as far away from lawyers as you or I do, by placing the onus on you for anything you may use with their service - makes sense. A text prompt is way less fraught with legal issues, and is what I’ll be using here.
So with 300 credits burning a hole in my pocket, it was time to let the machines help me get creative.
To the left of the window is a sidebar. Within it is the text prompt which will turn my words into digital magic. OK. Let’s start with ‘futuristic flying car’ - simple enough. Now of course I could go into much greater detail - in fact, a prompt can have up to 600 characters, but I was curious to see what a simple request would produce…
And the first attempt was a disappointment. Sure the car itself looked futuristic enough, but it had massive wheels, and wasn’t even floating. It’s clear I’ll need to be way more descriptive.
20 credits down, 280 to go.
My next prompt was “futuristic car with no wheels, with massive thrusters where the wheels should be, with wraparound windows from the front to the back.” What I got was just a car. With wheels. OK, we’re not understanding each other.
Let's try “futuristic flying vehicle that hovers above the ground, the passenger area is completely covered with a roof and windows.” OK, now we’re getting somewhere. At least the wheels are gone, and my flying car looks like a cross between a helicopter and a plane, minus the propellers. But I can work with that.
Actually I’m going to risk another 20 credits and see what else this AI can come up with using the exact same prompt by clicking on the ‘Regenerate’ button (the first in a row of three buttons at the bottom of the currently selected image).
Oddly enough, it gave me almost exactly the same design. OK, I still prefer the first one. That’s the cool thing about Flow Studio: every image you’ve asked to be created can still be accessed in the Library section.
Having selected my preferred vehicle, I’d like to make a few changes to it. This is where the ‘Edit’ button comes into its own. It allows you to alter the existing image rather than start the process from scratch. I asked it to “make the vehicle red with yellow stripes”. Now, “Change the yellow stripes to black, and make the red darker.”
Now I must confess to having made a mistake: excited as I was to see the first color change, I just typed in the prompt field, not realising I was in ‘new image’ mode, not ‘edit image’ mode. Both are located exactly in the same place and look identical aside from the menu’s title.
So, I wasted 20 credits, but it’s a lesson worth learning: once you've sent your prompt while in edit mode, the interface will take you back to the default “new image” prompt, so you need to keep your wits about you: check that sidebar for the tell-tale signs you’re working with the right tool, so you don’t needlessly lose credits - as I did.
Having now edited my previous image to the desired color, I'm left with 160 credits - nearly halfway through my allotment. These credits dwindle fast. Better move on to the next stage of the process.
Image to 3D
We explored the ‘Regenerate’ and ‘Edit’ buttons. The last one in that row is ‘Use as 3D input’. Let’s do that.
Clicking on that button takes you to the ‘Image to 3D’ section. The only main difference is that your image, which you can still see in the main preview section, has now been inserted in the ‘Image Prompt’ field, which has replaced the text prompt section, in the left sidebar.
Time to see what Flow Studio can do with my image. Let’s give our model a name (note: you can’t generate a 3D image without a name, and that’s not done automatically for you). Click on ‘Generate Mesh’.
The process understandably takes longer to perform than just creating an image, yet it costs the same: 20 credits.
We end up with 4 different yet almost identical versions, based on that same image. I’m going to click on the one with the most polygons. This leads me to a window with yet more options.
As I noted in my previous article, "you’re offered four different versions of the proposed 3D model, each with various numbers of faces (aka flat surfaces or polygons) and vertices (aka individual points in 3D space)."
Having set my own preferences for this exercise, it’s time to ‘Generate’ once more.
I have to admit the result is pretty good, considering how few prompts it took to get here. And I’d still have enough credits left to experiment with retexturing if I wanted.
So I could say, the answer to my original question is “yes”: 300 credits can be enough to create a 3D model from scratch. Time to export this creation.
Export
And it was all going so well…
It turns out that the limit here isn’t the amount of credits: there should be enough to create what you’re after, with a bit of experimentation - and some errors - thrown in. The biggest hurdle lies when it’s time to export your - or rather the algorithm’s - work.
You can’t.
Yes, there’s a ‘Download’ button, conveniently located lower right of the main preview section, and it offers you three formats, USD, OBJ, STL. But they’re all locked behind a paywall.
Any of the paid tiers will enable downloads it seems, but sadly that’s the end of the road for those on the free tier.
“Hang on, Steve,” I hear you say, “what about the fact it’s stated that the free tier has a ‘maximum export resolution of 720p’?”
Yes, that was a puzzle, I’m grant you that.
We may not be able to export 3D models, but maybe generated images are available? However, going back to one of those I’d asked to be created at the start of this experiment, you’ll find no ‘Download’ button anywhere.
However, that doesn’t mean you can’t save a copy outside of Flow Studio: simply right-click on the large preview version (not the thumbnail in the Library section on the right), and you’ll be given the option to save the image anywhere on your drive.
I have to admit, this is disappointing. Even just allowing one 3D download a month would’ve been better than nothing whatsoever. Still, now we know this, we can focus our precious credits on using the service to create 2D images.
It’s not the outcome I was expecting when I started this experiment. At least I didn’t come out of this empty handed.
Final Thoughts
Like all generative AI services, the output depends greatly on what you give it. What might seem obvious to you or I, likely won’t be to a machine, so you’ll have to work on your prompts until you start to think like it does. That’s not a criticism of Flow Studio at all - just how this technology currently works.
I found the results to be excellent. Actually, having some credits to spare, I went back to the original car it created for me, and asked it to remove the wheels via the ‘Edit’ feature. And the new version was much better. It even made it hover. I can download that image, just not any 3D model.
You can’t begrudge Autodesk for that though: this technology isn’t cheap. I just wish this limitation was clear when looking at the various subscriptions.
At least you won’t lose any of your work: when you’re ready to pay for a subscription, any projects you created while using the free tier will be there, ready for you to download or work on further.
