#6: Launch a SaaS like Meshy or Rodin next week with Open Source
Dive into the growing 3d model generation market, and learn how to build your own using open source tech!
Most successful SaaS products, like Slack, Shopify, Zoom, Dropbox, and HubSpot, weren’t first of their kind; They didn’t invent their fields - they just made existing ones better.
Author’s note: Reach out to me on Substack if there's a product or topic you'd like me to explore in future posts. Shoutout to Oryamon for suggesting this one!
What are 3d generation tools?
They are AI-powered tools that use text prompts or images to generate detailed 3D models. These models can then be used as assets across creative fields such as gaming, animation, architecture, and more. Most 3d generators give you fine control over your design (eg. realistic vs stylized) and usually include complementary features based on their industry focus (eg. rigging and animation for game devs and animators).
Let's look at the market!
The natural next phase after image and video generation, research in 3D model generation has been steadily picking up pace and is becoming a more feasible option for creators. With market interest at an all time high (source), AI-powered 3D model generation is projected to become a major disruptor across creative sectors:
Keywords like “ai 3d model generator” and “image to 3d model” presently get between 10k to 100k monthly searches with low competition (source: google keyword planner) and a broad target audience - spanning across gaming, VR/AR, Film & animation, architecture, and other design disciplines.
With very few competitors in this fresh market, Meshy.ai ('21) has grown to 1.6M ARR, while Hyper3D ('20) just raised an undisclosed series A with Bytedance (TikTok parent) as their lead investor. They have usage based pricing, with standard plans ranging roughly between $25 to $100 dollars.
Alright, so how do we build this quickly?
3D generation tech is evolving, but it’s still tricky. 3D data is way more complex than 2D (eg. images, and 1D text), needs heavy hardware, and the available training data is magnitudes smaller compared to text and images. Since the field is still new, there’s no clear “best” method, but most models use some form of an encoder-decoder setup. Here’s how the impressive, and very new, Trellis and Hunyuan3D models work:
The encoder take inputs (like text or images) and turn them into a latent (hidden/intermediate) 3D representations. Trellis uses a sparse LEGO®-like representation where each block (voxel) holds a latent vector for shape and texture, while Hunyuan3D uses a diffusion model (followed by a ‘variational auto-encoder’) to create the latent 3D shape.
After encoding, decoders turn the latent representations into full 3D assets - which could be a mesh or a radiance field, depending on the use case. Both Trellis and Hunyuan3D create realistic textures by cleaning up lighting and shadows.
Here are the official open source projects on GitHub:
TRELLIS by Microsoft
Hunyuan3D-2 by Tencent
Worried about building signups, user management, payments, etc.? Here are my go-to open-source SaaS boilerplates that include everything you need out of the box:
SaaS Boilerplate by Remi Wg
Open SaaS by wasp-lang
Launching soon? DM me if you’re looking for a high-impact landing page, sales deck, or any other GTM collateral to get your product in front of the right people.
How will my SaaS stand out in the noise?
The relatively fresh market makes it easier to differentiate on core principles and achieve product market fit:
Focus on specific industries: 3D models aren’t just for gaming. Design and tailor your product to solve challenges for specialized areas like architecture, film and animation, product design, XR, etc.
Add unique features to increase switching cost: Identify gaps raised by the 3d community to build sticky features such as niche 3rd party integrations, legacy file formats, assistive and collaborative tooling, unique pricing models and more.
TMI?
I’m an ex-AI engineer and product lead, so don’t hesitate to reach out with any questions!
P.S. I started this free weekly newsletter to share open-source / turnkey resources for recreating popular products (like this one). If you’re a founder looking to launch your next product without reinventing the wheel, please subscribe :)