AI Underground / Atlanta / 2026

GIVE IN
TO AI.

Story, identity, automation, and the uncomfortable power of letting machines carry the load.

MAX THOMAS · AI STORYTELLER · TECHNICAL ARTIST

Portrait of Max ThomasAI UndergroundMAX THOMAS
AI STORYTELLER / TECHNICAL ARTIST

Opening position / proven under production pressure

I AM NOT HERE
TO DEFEND AI.

I am here to show what happens when you build your life around it.

FILM-SCALE PRODUCTION · PROGRAM BUILDING · PUBLIC AI WORKFLOWS

Performance capture stage work for Moana
MOANA / PERFORMANCE CAPTURE
Virtual production at Cine Gear
VES / CINE GEAR / VIRTUAL PRODUCTION
ActorCapture workshop in South Korea
ACTORCAPTURE / KOREA WORKSHOP

Max Thomas / public work

AI
STORYTELLER.

Brief. Cast. Stage. Generate. Finish. The same production discipline, rebuilt around models that keep changing.

ActorCapture / CTO / production systems

THE TASKS MOVED.
ONE BY ONE.

01 / OBSERVEFind repetitionWatch the small tasks that already loop.
02 / ENCODESet the rulesTurn judgment into criteria and constraints.
03 / ROUTEChoose the toolMatch each task to the right model.
04 / ITERATELet AI carry itDraft, test, correct, repeat.
05 / FINAL GATEI reviewI still own the final output and release.

At ActorCapture, I watched iterative work migrate to AI one small task at a time. The labor moved. Accountability did not.

Coyote Crunch / local → cloud

QUALITY
BROKE THE TIE.

Coyote Crunch started from a disciplined 3D production base. Cloud video eventually crossed the threshold where the finished motion justified changing the infrastructure decision.

Use when cinematic motion quality is the deciding variable.

The trade: less infrastructure control, more finished-frame value.

Coyote Crunch commercial public Reallusion featureCOYOTE CRUNCH / PUBLIC COMMERCIAL FEATURE

Governance / efficient model distribution

THE RIGHT MODEL.
THE RIGHT JOB.

Established rulesApproved contextTool permissionsProvenance + review
01 / ROUTINEEfficient modelsRetrieve. Organize. Summarize. Draft.
OllamaQwenGemma
02 / SPECIALISTExact production tools3D. Real-time. Code. Media. Documents.
iCloneBlenderUnrealComfyUI
03 / FRONTIERQuality modelsEscalate for image, motion, reasoning, and finish.
GPTGeminiClaude
04 / HUMAN GATEMax reviewsCorrect. Approve. Own the release.Max Thomas

Efficiency chooses the default. Quality triggers escalation. I decide what leaves the system.

Sidewalk Syndicate / recurring cast across shots

CHARACTER CONSISTENCY
IS PRODUCTION DESIGN.

Sidewalk Syndicate recurring cast in daylight
DAYLIGHT / GROUP
Sidewalk Syndicate recurring cast in a wide shot
WIDE / ACTION
Sidewalk Syndicate recurring cast in close coverage
CLOSE / COVERAGE
Sidewalk Syndicate recurring cast in dramatic night lighting
NIGHT / DRAMA

Same faces. Same silhouettes. Same world. Different shots, lenses, and lighting conditions.

GEOMETRY · WARDROBE · PERFORMANCE · CAMERA · MEMORY

Published workflow advisor / traditional structure, new tools

THE PROCESS
DID NOT DISAPPEAR.

I advise tool builders by mapping AI into a production pipeline artists already understand. Lock what must remain consistent. Spend uncertainty only where it adds value.

01Brief
02Previs
03Production
04Finish

Character Creator and iClone preserve the structure. Generative models expand the finish.

Sidewalk Syndicate 3D foundation and shot setup
3D FOUNDATION / CONTROL
Sidewalk Syndicate finished generated frame
GENERATIVE FINISH / VALUE

CMIS 6385 / virtual cinematography / course system

TEACHING BECAME
A PRODUCTION SYSTEM.

I did not automate away the teaching. I built a governed environment around it: RED Komodo, Unreal, OptiTrack, ICVFX, branded iCollege modules, production briefs, Discord operations, schedules, grading, and student communication.

MORE HUMAN DIRECTION · STRONGER STRUCTURE · FASTER ITERATION · ACCOUNTABILITY

Focused CMIS 6385 Komodo Lab course interface
KOMODO LAB / CAMERA ANATOMY
Focused CMIS 6385 virtual cinematography production monitor
VIRTUAL CINEMATOGRAPHY / PRODUCTION DATA

The consequence

I DIDN'T AUTOMATE
MYSELF AWAY.

I moved farther and faster than the institution knew how to manage. The work arrived before the structure around it was ready.

THE LIMIT WASN'T THE TEACHING. IT WAS INSTITUTIONAL READINESS.

Why every presentation became HTML

THE DECK
BECAME SOFTWARE.

Responsive instead of fixed. Interactive instead of linear. Searchable, linkable, versionable, deployable, and fast enough for an agent to revise while the conversation is still happening.

This presentation is the proof.

public Reallusion workflow deck / live HTML
Complete frame from a published Reallusion workflow presentation
ONE COMPLETE FRAME / REAL PUBLISHED HTML DECK

The real Max-Life-OS graph / markdown / governance / retrieval

OBSIDIAN IS
THE SOURCE OF TRUTH.

Actual Max-Life-OS Obsidian knowledge graph
01 / FILESYSTEMDurable markdown
02 / GOVERNANCENames, ownership, state
03 / PROVENANCESource, date, context
04 / RETRIEVALAgent-ready memory

Alibaba Cloud + Codex + APIs + MCP

ONE PERSON.
A CONNECTED STUDIO.

ObsidianGoverned knowledge
Files + webEvidence and assets
Human directionIntent and approval
Hermes
Agent

PLAN / ROUTE / EXECUTE / VERIFY

CodexCode and systems
Alibaba CloudCloud models and services
API + MCPApps become tools
AI and real-time production pipelineCAPTURE → 3D → MODELS → DELIVERY

Complete production pipelines

PRODUCTION
IS A GRAPH.

A brief can create tasks. Tasks can call code. Code can operate models. Models can return media. Media can enter review. Approval can trigger delivery.

The value is not the API call. It is the governed chain of handoffs.

Black Sand Tees / new business

I AUTOMATED
THE INVENTORY.

Kona Night Manta shirtNIGHT MANTA
Kilauea shirtKĪLAUEA
Outrigger shirtOUTRIGGER

Design + inventory + marketing front end

ONE PIPELINE.
A WHOLE BUSINESS.

01Design system
02Mockup inventory
03Storefront
04Campaign media

The same source artwork feeds product variants, collection logic, sales visuals, location stories, and a responsive marketing surface.

Black Sand Tees automated storefrontBLACK SAND TEES / LIVE FRONT END

Live proof / Max + Joey + ChatGPT + Suno

I MADE THIS
WHILE I WAS HERE.

I dumped in an idea about HODLBarbarian. ChatGPT drove the lyrics and style prompts. Joey helped shape the moment. Suno made it audible.

HODLBARBARIAN / SPEAKER PROFILE ↗

Hermes Agent / trained skills

TASTE CAN BECOME
INFRASTRUCTURE.

A good skill does more than store a prompt. It encodes the brief, sequence, tools, constraints, failure modes, verification, and delivery standard.

The agent stops guessing how you work.

# songwriting-and-ai-music trigger: idea needs emotional form input: message + audience + tone workflow: 01 shape the lyric 02 design the sound 03 generate variants 04 judge against intent 05 deliver + archive standard: human meaning survives

Live proof / intelligence becomes infrastructure

ALIBABA CLOUD GIVES ME SCALE.
HERMES GIVES IT A SYSTEM.

Alibaba CloudFrontier intelligence
Model distributionLocal, specialist, frontier
Scalable servicesInference, storage, APIs
HERMES
AGENT
CONTEXT · MEMORY · SKILLS · TOOLS
PLAN · ROUTE · EXECUTE · VERIFY
Max-Life-OSGoverned working memory
Connected toolsCode, browser, desktop, media
Working artifactsNot answers. Outcomes.
HUMAN RELEASE GATEMax sets intent, corrects the work, approves the output, and owns the result.

The actual proposition

BUILD IN PUBLIC.
GOVERN IN PRIVATE.

Use AI where it compounds you. Keep identity, provenance, approval, and purpose under human control.

MAX THOMAS · AI UNDERGROUND · ATLANTA · 2026

QR code for this deck

SCAN FOR THE LIVE DECK