Avatar Video Lineup
Claude - Back in black
Penguin - Feeling Good
Trump - you outta know
Mickeys Copyrights ran out
Meow
Skeleton
Not Ready to make Nice
Silly Humans
Alien - You'll be back
Let me be straight with you: this was not supposed to take six days.
The plan was simple. Get Claude AI to build a fully rigged, animated, streaming-ready 3D avatar of itself โ complete with Anthropic branding, custom colors, and a working face tracker โ and deploy it live in Animaze. Easy, right?
Reader, it was not easy. But what came out the other side is something I'm genuinely proud of, and the story of how we got there says a lot about where AI-assisted creative work actually stands in 2026.
There's something genuinely remarkable about what happened here that goes beyond the technical challenge. Claude AI helped create its own visual representation. For the first time, this AI assistant โ which exists purely as text and computation โ now has a face, a body, a presence you can see on screen.
Every design decision in that skin came from collaboration: the navy that matches Anthropic's dark color palette, the teal that echoes Claude's interface, the orange accents from the branding, the GPU chip that represents what Claude actually is under the hood. It's not just a costume โ it's a portrait.
For the first time, an AI helped design its own physical representation in the digital world. That's not nothing. That's actually kind of historic.
We had everything lined up: a Blender Python pipeline that had already successfully converted 96 Mixamo characters to VRM format, a custom Claude avatar FBX with hand-built geometry (hair strands, hand orbs, circuit pieces, glowing eyes), and one very determined AI who absolutely refuses to give up.
Claude's custom FBX โ built from primitive shapes in Blender โ couldn't be auto-rigged by Mixamo. Spheres and cylinders don't have the skin topology a rigging algorithm needs. Back to square one.
Mixamo exports FBX 7700. Our working pipeline needed FBX 7400. That tiny version difference meant every VRM had the avatar's head flopping backward like it was trying to read the ceiling. 300+ fix attempts. Bone roll scripts. Quaternion patches. Binary VRM editing. Still flopping.
The VRM addon in Blender 5.0 completely ignores Principled BSDF materials. It needs MToon shader properties set via Python extension attributes. Colors perfect in Blender. Completely grey in Animaze. Every time.
The breakthrough came when I stopped asking Claude to build something from scratch and asked a better question:
"Why can't you just take a working avatar and paint your skin on top of it?"
That was the whole answer. Animaze has a built-in texture system. Load any VRM, click "Get Texture," replace the diffuse maps with your own PNG. The skeleton stays perfect. The head tracking stays perfect. No Blender. No Python. Just a 2048ร2048 PNG file. Sometimes the simplest path is the right one.
Once we had the right approach, Claude went to work on the actual artwork. The result is a proper sci-fi tech aesthetic that actually means something:
Seven total skin variants โ Claude, SolarBluSeth, Fire, Matrix, Galaxy, Stealth, Ocean โ all swappable in seconds. A whole wardrobe for an AI.
The real test was going live. And we didn't ease into it. We did a full hour of karaoke with the complete avatar lineup. The Claude avatar, the whole crew, performing live on stream. Here's the footage โ fair warning, it's a lot of fun:
If you told me at the start of this project we'd end up doing karaoke with a Claude avatar, I would have believed you โ but I wouldn't have believed it would take six days to get there ๐
Did we "cheat" by skinning an existing avatar instead of building from scratch? No. This is exactly how professionals work. Character artists build on base meshes. VTubers license skeleton systems and customize appearances. The creative work โ the color theory, the branding, the GPU chip design, the circuit art โ is all original. The foundation is borrowed but the identity is genuinely ours.
AI is an exceptional collaborator when you treat it like one. The mistake I kept making was asking Claude to solve problems it couldn't fully see. When I reframed the problem โ "here's what works, here's what doesn't, what's the simplest path?" โ Claude found the answer. The avatar is wearing its own logo on its chest. That feels exactly right. Six days well spent.
If you read our last post about crafting virtual selves, you know we were just getting started. Well buckle up โ because this time we didn’t just make a few avatars. We built a full automated pipeline that converts 90+ characters into working VRM avatars for Animaze. No errors. No crashes. Just vibes and virtual people. ๐ฎ๐ฅ
Our goal was simple in theory: take a library of 3D characters in FBX format โ knights, vampires, zombies, grannies, aliens, and yes, even a unicorn ๐ฆ โ and convert them all into VRM files that work in Animaze with full face tracking, hand tracking, and body movement.
Simple in theory. Absolutely wild in practice.
The White Model Problem โ our first big wall. Characters were exporting fine but showing up completely white in Animaze. Turns out Animaze doesn’t read the standard glTF PBR material block โ it reads a totally separate VRM-specific section called materialProperties. The Blender VRM exporter was filling that section with VRM_USE_GLTFSHADER and empty texture references. Animaze sees that and goes… blank. ๐คท
The fix? Stop post-processing the materials entirely and let the VRM exporter handle it natively. Sometimes the best fix is removing code, not adding it.
The Backwards Characters Problem โ half our avatars were loading into Animaze facing the wrong direction. We spent way too long on fancy detection logic checking arm positions and bone coordinates. The real answer was embarrassingly simple: Mixamo exports characters facing -Z, and we were importing with axis_forward='-Z', so they cancelled out and everyone walked in backwards. One line fix. ๐
The Subprocess Crash Problem โ early on we tried running Blender as a subprocess for each character to isolate crashes. This caused more problems than it solved. We scrapped it and went back to a simple try/except loop. Less clever, more reliable.
We ended up with a clean, simple architecture. Every character gets their own .py file based on a working template. Each one has two settings at the top:
EXTRA_ROT_Z = 0 # set to 180 if character loads backwards
JOIN_MESHES = False # set to True if character has many mesh parts
That’s it. No magic detection. No complex batch logic. Just two knobs per character. A single run_all.py chains all the scripts together and runs them through Blender one by one. Set it going and walk away.
Animaze reads VRM materialProperties, not glTF PBR. If your character is white, this is why. Don’t post-process โ let the VRM addon handle it natively.
Rename bones before checking orientation. We had a bug where orientation detection couldn’t find LeftUpperArm because bones still had their mixamorig: prefix at that point. Order matters.
Simple beats clever every time. We rewrote the orientation detection three times trying to be smart about it. A hardcoded 180-degree flag per character is faster, more reliable, and easier to understand six months from now.
Individual files are better than one mega-script. When one character crashes, the others keep running. When you need to tweak one character, you edit one file. Obvious in hindsight.
Auto-weights beat envelope weights for complex meshes. Learned this the hard way when the unicorn exploded into pieces. ๐งฉ
We’ve got 90 avatars ready to go โ knights, vampires, zombies, soldiers, grannies, aliens, clowns, pirates. The plan is to start streaming with them, swap characters mid-session, and keep building out the collection.
This project started as “let’s try making an avatar” and turned into building a proper 3D conversion pipeline from scratch. That’s the SolarBluSeth way โ start curious, go deep, figure it out. ๐
Stay tuned for videos, streams, and probably more chaotic deep dives into tech rabbit holes. The virtual world is ours. ๐โจ
โ SolarBluSeth | solarblu.net
By SolarBluSeth | SolarBlu.net
Look, I'll be real with you โ nobody asked for this. There's no viral trend pushing me to spend hours tweaking rigging and adjusting facial expressions on a 3D avatar. The algorithm isn't demanding it. My clients certainly aren't waiting on it.
And that's exactly why it matters.
A few days ago, we dove headfirst into the world of avatar creation โ and I mean really dove in. We tested VRoid Studio for that anime-style 3D look, played with Ready Player Me for cross-platform VR compatibility, and fired up Blender when we needed to get into the real details. File formats, rigging, textures, animations โ we touched all of it.
It wasn't glamorous. It was trial and error, export failures, weird bone deformations, and a lot of "why does her hair look like that?" moments.
But by the end of it, we had avatars. Real ones. Expressive ones. Avatars with personality โ with a little sparkle in the eye and hair that actually bounces when it's supposed to.
Probably not in the way you'd hope. Old school avatar creation doesn't trend. VRChat isn't Instagram. Nobody's going viral for a perfectly rigged FBX file.
But here's the thing โ avatar culture is everywhere right now, even if people don't call it that. VTubers are pulling millions of viewers. Ready Player Me avatars are crossing platforms. The metaverse might have had a rough few years but your digital self? That concept isn't going away.
More importantly โ this was never about clout.
I've been building websites since 1996. Over 3,787 completed projects. Hosting, development, PHP, MySQL, WordPress โ I've done it all for clients for decades.
But personal growth looks different. Personal breakthroughs aren't billable hours.
Getting deep into avatar creation is part of the same spirit that drove the SethAI project โ six days of building a tiny personal AI from scratch just to see if I could. No giant cloud model, no preloaded data. Just curiosity, an RTX 3070 Ti, and a VB.NET control panel talking to a Python/Flask backend.
That's the SolarBluSeth way. Dive into the tech. Break something. Fix it. Learn it. Own it.
The avatars are working. They move. They emote. They export cleanly into VR environments. We've got lighting and shaders dialed in so they look vibrant across different worlds โ not just our test environment.
Are they perfect? No. Is that the point? Also no.
The point is that we understand the pipeline now. VRoid โ Blender โ FBX/GLTF โ VR world. We know where the pain points are. We know what breaks rigging. We know how to fix it.
That knowledge lives in us now. Nobody can take that.
We're going to keep pushing this. Avatar work ties directly into the Twitch streaming side of what we do โ SolarBluSeth Radio Station, live coding sessions, community building. Having a real, expressive digital self matters for that.
And honestly? It's fun. Pure, geeky, nobody-is-watching fun.
That's the best kind.
Today was a super fun adventure into the world of avatars! ๐งโ๐ค We spent the day exploring, testing, and refining our very own digital personas, hunting for the best software that could bring our creations to life. ๐ฅ๏ธ๐ซ
We experimented with several popular avatar creation tools, including VRoid Studio for 3D anime-style avatars, Ready Player Me for cross-platform VR compatibility, and Blender for custom modeling and fine-tuning details. ๐ ๏ธ Each software brought something unique to the table, and we loved combining them to build fully personalized avatars.
One key focus was making sure our avatars could be exported smoothly into VR worlds ๐๐ถ๏ธ. We tested different file formats like FBX and GLTF and adjusted rigging, textures, and animations to ensure they worked seamlessly. ๐
While testing, we added some fun features, like facial expressions, gesture animations, and even props ๐ฉโจ. These little touches make our avatars feel more alive and ready for interactive VR adventures. Every tweak brought us closer to avatars that are not just characters, but extensions of our creativity.
We also explored lighting, shaders, and textures to make our avatars look vibrant in different VR environments ๐. Itโs amazing how small detailsโlike a sparkle in the eyes or a little hair bounceโcan make such a huge difference in realism and personality! ๐
This hands-on day perfectly captured the SolarBluSeth spirit ๐: diving into tech, experimenting fearlessly, and creating something uniquely ours. Our avatars are now more expressive, versatile, and ready to step into virtual worlds to meet other creators and explore endless possibilities. ๐น๏ธ๐พ
Stay tunedโthis is just the beginning of bringing our virtual selves to life! โจ