404-GEN

Overview / The Problem

404-GEN is a technology studio building the largest open-source library of AI-generated 3D models, with over 6 million assets. Their Blender add-on gave 3D artists and game developers direct access to that library -- but conversion was low. The product had real capability and no audience.

The stakes were higher than a typical launch. 404-GEN's revenue ran through Bittensor, a decentralized AI network where user adoption directly drove model generation and network rewards. Low adoption wasn't just a growth problem. It was a revenue problem.

I was brought in as a 3D design consultant to close that gap:

• Design and launch storefronts across Gumroad, Unity Store, and Blender Market
• Produce marketing content to drive awareness and adoption
• Explore new use cases to identify angles that would actually convert

404-GEN website: 404.xyz SN17 whitepaper: doc.404.xyz/sn17-whitepaper

The User

404-GEN had two distinct user groups that needed to work in parallel.

Primary Users: 3D Artists and Game Developers. The core audience was 3D artists and game developers looking to cut the time and cost of asset creation. For them, the value proposition was straightforward: a free, AI-generated 3D model library that slotted directly into their existing Blender, Unity, or XR pipelines. These were the users who needed to be sold on the product. Without their adoption, the system had no purpose.

Secondary Users: Miners and Validators. Behind every model was a decentralized network of contributors:

• Miners ran the compute that generated 3D models from text prompts
• Validators assessed each model's quality, scoring against accuracy and visual fidelity before it ever reached an artist

This created a self-sustaining incentive loop -- more artists using the product meant more demand for models, which rewarded miners and validators for higher quality output. Adoption at the front end kept the entire system healthy.

Launch Design

The core design challenge wasn't the product itself -- it was credibility. 3D artists and developers are skeptical audiences. Showing a model in a void wasn't going to convert anyone. The product needed to be seen in context, doing real work, inside environments that felt familiar to the people we were trying to reach.

The direction we committed to: use the product to market the product. Every asset in every piece of marketing content was generated with 404-GEN. No stock imagery, no placeholder scenes. The models appeared inside actual Blender setups, game-ready environments, and XR scenes -- proving the use case by living inside it.

This informed everything we launched:

• Gumroad and Blender Market storefronts - designed around in-context scenes that showed the product working inside real creative pipelines
• Unity Store page - framed around game developer workflows, leading with what a developer could build, not what the technology was
• Discord bot - streamlined the generation flow so artists could go from prompt to model without leaving their community space
• Tutorial and installation videos - built to reduce activation friction, walking users through setup and first use with the same in-context visual language

Public Releases

With conversion stalling on the Blender add-on, the team needed new angles. The question wasn't whether the technology worked, it was whether artists and developers could see themselves using it. My role shifted into active exploration: stress-testing the product across different creative contexts to surface use cases that hadn't been marketed yet.

The approach was hands-on. I worked directly in Blender, pushing the models beyond their obvious applications to find where they had untapped potential:

• XR and immersive environments - testing Gaussian splats inside spatial contexts to evaluate feasibility for AR/VR pipelines
• Geometry node animations - leveraging the point-cloud structure of splat models to create fluid, morphing animations that traditional meshes couldn't replicate
• Scene composition - building full environments using only 404-GEN assets to identify which model types held up under real production conditions

The goal was to find proof points that would resonate with a developer or artist who hadn't been convinced yet. Findings fed directly back into the marketing direction -- new content angles, new audiences to speak to, and a clearer picture of where the product had room to grow.

What shipped

Across six months, the work moved from exploratory prototypes into public releases and a clearer path for artists to try the product.

• About 40 videos produced; 10–15 used in official marketing campaigns.
• Three storefronts launched: Gumroad, Blender Market, and Unity Store.
• 404-GEN became a Unity Verified Solution during the engagement.
• The splat-to-UV-mesh workflow was adopted by the product team.
• Optimization findings from hands-on tests fed into the generation pipeline.

The work followed one principle: show the product doing real work in the hands of artists and developers.

Launch films

A closer look at the models, the tools, and how the work was introduced to artists.

Scenes built with 404-GEN

Generated assets move into environments, compositions, and visual worlds.

Form and material studies

Short experiments in texture, motion, and everyday objects.

See more videos on Instagram↗

What I learned

I learned the most by putting the models into real scenes. Storefronts and tutorials helped people see what they could make, but testing the assets myself also exposed rough edges in the workflow. One of those experiments, converting splats to UV meshes, made it into the team’s pipeline. I’d start there again: make something with the tool, see where it breaks, and show the result honestly. My computer crashed more than once while testing the scenes, which was a useful reminder that the workflow still had limits.

Thank you to the 404-GEN Team

Ben James

Monica Vasandani

Ruben Hohndorf Maxim Bladyko

Alexander Tereshin

Rodion Stetsurin

“Sheesh”

Robin Simons