Media Creation Curriculum in the AI Era #1 — Andrew Price

Updated: 2026-05*

1. Introduction

This essay is a Q&A between the author and Claude AI about the author’s vision for redesigning a media-production course, and it will continue across multiple installments. This first installment is built around Andrew Price’s remarks. Since 2023, AI image and video generation has advanced fast enough that you can no longer build a media-production course on the same assumptions you used five years ago. With Tokyo Metropolitan University’s “Visual Practicum I/II” in mind, the goal here is to walk through outside debates, real-world cases, and adjacent tools, and from that derive a curriculum design that should hold up over a multi-year horizon.

For this round, I take Andrew Price (Blender Guru) — best known for the Blender donut tutorial — and what he said about “the future of AI and 3DCG” across 2025–2026, and use that argument as a starting point for the course redesign. Price’s position is not the simple “3DCG is dead” line; it carries the nuance of “redefining 3D as a control surface for AI.” That distinction is worth reading carefully.

1.1 References

References:

1.2 What this piece covers

A summary of Price’s argument on AI and 3DCG
What remains in 3DCG, what disappears
Production techniques likely to survive the next 5–10 years
A redesign proposal for “Visual Practicum I/II” (180 min × 15 sessions × 2 semesters)
Risks and caveats

2. Andrew Price’s argument

I’ll organize Price’s output along three axes: the overall picture, skills that stay vs. skills that fade, and his practical proposal.

2.1 The overall picture

Price’s central claim is that AI is shifting the locus of creative work from large studios to individuals and small teams. At SIGGRAPH 2025 he emphasized the democratization fact: real-time path tracing inside Blender, NVIDIA DLSS, and other capabilities that used to be locked inside large studios now run on individual machines. As he frames it, we have reached the point where three people can make a film at the level a large studio used to.

His position is that what AI threatens is not individual artists but large enterprises burdened by scale and cost. The artists who will be paid in this new world are what he calls “high-agency” creators — those who can not only conceive a story but autonomously finish a film or a game by themselves. AI does not have to beat Disney or Pixar’s perfect hair simulation; it just has to win the way YouTube won against late-night TV — through niche, authentic stories.

2.2 What stays, what fades in 3DCG

Price’s own LinkedIn post is the clearest summary. The gist:

Ask 100 artists to design “a vehicle on an imaginary planet” and you get 100 different answers. How do you train a model with ground truth on creation that has no ground truth?
For the parts of creativity that text input cannot control, 3D gives you much finer control — so long careers as a 3D artist remain viable
That said, some skills do lose value under specific AI workflows
Losing skills you spent years acquiring is no fun, but that is hardly new
Believing you can resist AI workflows forever on the back of someone else’s economic pain is, frankly, foolish

2.3 “Use 3D as a control surface for AI”

This is Price’s central practical proposal as of 2026. In his BCON 2026 (the North America edition held in Austin) talk, “How to control AI with Blender,” he positions Blender not as a modeling app but as a control tool for instructing AI.

The workflow he presents can be summarized like this:

Block out rough spatial composition in Blender
Export a depth pass, hand it to ComfyUI, and steer the AI image generation in a trustworthy direction
Build a 3D placeholder from the generated image
Construct an AI-assisted “worldbible” of setting and style
Batch-generate 3D assets and combine them Lego-style

Two phrases from this section matter.

The first is “Labor vs. judgment.” What AI replaces is the labor — what remains for the human is judgment: taste, composition, world-building, choice of story.

The second is the warning “Automate only the skills you’re willing to lose.” The implication is that choosing what to automate is not a technical decision but an identity decision.

3. Production techniques that survive the next 5–10 years

Generalizing Price’s argument — and discounting his Blender bias — production techniques likely to survive the next 5–10 years sort roughly like this.

3.1 What stays — or gains value

Story construction, world-building, character setting documents (“worldbible”)
Judgment in camera language, composition, editing, timing
Performance, physicality, live-action material — the act of making something happen in front of a camera
The ability to direct AI: prompting, references, depth and pose as structural constraints
Copyright, ethics, and dataset judgment

3.2 What gets automated

First-pass generation of individual assets (modeling, texturing, rigging)
In-betweens, lip-sync, simple VFX
The first few dozen rough concept-art passes
First-pass color correction and compositing

3.3 The gray zone

3DCG modeling and animation will shrink as final deliverables, but may grow in demand as blueprints and intermediate representations to control AI generation (Price’s depth-pass workflow is the canonical example)
Real-time node-based environments like TouchDesigner have room to be re-evaluated as the output stage / live-performance stage for AI-generated material

4. Direction for the curriculum redesign

From here on, the discussion turns to concrete course design.

4.1 Current state and draft proposal

The class-time budget:

Visual Practicum I: 180 min × 15 sessions (first semester)
Visual Practicum II: 180 min × 15 sessions (second semester)
Total: 30 sessions / 90 hours

Current composition and the draft under consideration:

Current: Blender 60% / After Effects 20% / TouchDesigner 20%
Draft: Blender 0% / Comfy Cloud 20% / Runway 60% / live-action shooting (action cameras etc.) 20%

4.2 Evaluating the draft

The direction is right. Anchoring on live-action, consolidating character design via image generation, then extending into video generation — that pipeline aligns with current AI strengths (image-to-video, video-to-video, pose transfer). The idea of converting from performance also matches AI’s “respect the source material” mode.

Two points are worth reconsidering, though.

First, “Runway 60%” — that level of single-vendor dependency is excessive risk on a 5–10 year horizon. AI video generation is a field where the lead changes every six months; which of Runway / Veo / Sora / Kling / Wan / Hailuo ends up dominant is not yet settled. The course should be designed around pipeline patterns, not tool names.

Second, taking 3DCG to zero contradicts Price’s most recent line of argument (“use 3D as a control surface for AI”). Rather than zero, it is worth keeping a minimal slice — not modeling education, but 3D as a way to communicate space and structure to AI.

4.3 Revised ratio

Given the 90-hour total, here is the adjustment proposal. The ratios indicate the share of activity time across the course as a whole, not a strict per-session breakdown.

AI image / video generation (Comfy Cloud / Runway etc. consolidated): 50% (about 45 hours)
Live-action shooting and capture: 25% (about 23 hours)
3DCG (minimal “AI-control” configuration): 10% (about 9 hours)
Editing / judgment / direction practice / final production: 15% (about 13 hours)

Total 100%. By not pinning specific vendor names in the ratio, the course design survives tool churn.

4.4 Visual Practicum I (first semester, 15 sessions) — proposed structure

The first semester is centered on “asset creation and mastery of the first-pass pipeline.” For Price’s notion of “judgment” to become a real exercise, students first need hands-on experience with how the various assets are made.

Session 1: Orientation / overview of media production in the AI era
- Introduction to the “Labor vs. judgment” frame
- A bird’s-eye view of the current toolset (AI image / video generation, live-action, 3DCG)
- Learning goals and evaluation criteria
Session 2: Designing character and world (introduction to the worldbible)
- What a setting document is, and why it matters more in the AI era
- Exercise: assemble a one-sheet setting document from text + reference images
Session 3: AI image generation fundamentals ①
- Concepts of Comfy Cloud / ComfyUI; the minimum workflow
- Prompt structure, negative prompts
Session 4: AI image generation fundamentals ②
- Working with reference images (image-to-image, references, ways of specifying composition)
- Exercise: produce multiple images of the same character / same world
Session 5: Live-action shooting fundamentals ①
- Action cameras and related gear; minimum composition and lighting
- Exercise: shoot several short takes
Session 6: Live-action shooting fundamentals ②
- Directing performance, building a single shot, basic audio capture
- “Shooting with the downstream AI pass in mind”
Session 7: AI video generation fundamentals ①
- text-to-video — operation and limitations
- Seeds, camera specification, handling of duration
Session 8: AI video generation fundamentals ②
- image-to-video operation
- Exercise: animate the still from session 4
Session 9: AI video generation, applied
- video-to-video operation (transforming live-action footage)
- Exercise: transform the live-action material from session 6
Session 10: Editing fundamentals ①
- Introduction to an editor (DaVinci Resolve or similar)
- Cuts, tempo, connection
Session 11: Editing fundamentals ②
- Syncing audio and video; minimal sound design
- Minimal color grading
Session 12: First-semester assignment ① (planning, storyboarding)
- Plan a 15–30 second short
- Lock the shooting / generation plan
Session 13: First-semester assignment ② (asset generation, shooting)
Session 14: First-semester assignment ③ (editing, finishing)
Session 15: Critique and reflection

4.5 Visual Practicum II (second semester, 15 sessions) — proposed structure

The second semester is centered on “character consistency,” “AI-control via 3D,” and “finishing a piece.” Since individual skills have been picked up in the first semester, the second semester emphasizes integrating them into a single completed piece.

Session 1: Orientation / first-semester review and second-semester goals
- Recap of first-semester work
- Three axes for the semester: consistency / 3D control / finishing a piece
Session 2: Character consistency basics
- Techniques for keeping the same character across multiple shots (reference image, LoRA, IP-Adapter as concepts)
- From setting sheet to per-shot generation
Session 3: Performance-to-conversion pipeline ①
- Exercise: shoot performance and convert into a different character / style with AI
- pose-to-video, video-to-video, applied
Session 4: Performance-to-conversion pipeline ②
- Integrating performance and generated character
- Handling failure modes in facial expression, mouth, hands
Session 5: 3D as an AI control surface ①
- Minimal Blender operation (object placement, camera, lighting)
- Note: do not enter polygon editing — focus on space and composition
Session 6: 3D as an AI control surface ②
- Exporting depth / normal passes
- Exercise: hand the exported image to AI to control composition
Session 7: 3D as an AI control surface ③
- Exercise: generate multiple shots in the same space
- Sequence generation using 3D-side camera moves
Session 8: ComfyUI node-based construction (optional deeper dive)
- Building custom workflows
- Basics of batch generation
Session 9: Copyright, ethics, datasets
- The current debate around training data
- Cautions for publication / submission
Session 10: Second-semester assignment ① (planning, scenario, storyboard)
- Plan a 1–3 minute piece
- Lock the production schedule
Session 11: Second-semester assignment ② (asset generation, shooting, 3D design)
Session 12: Second-semester assignment ③ (rough edit)
Session 13: Second-semester assignment ④ (finishing)
Session 14: Critique preparation and screening check
Session 15: Final critique and wrap-up

4.6 Decision on the 3DCG slot

Three options compared:

Option A: take it to zero. Provide reference materials only
Option B: introduce it in the second semester across three sessions (about 9 hours) as “3D for controlling AI.” Limit to three points: blocking, depth-pass export, camera placement. Do not touch polygon editing
Option C: keep traditional foundational modeling across both semesters

Recommended: Option B. Three reasons.

First, as Price points out, the essential value of 3D is being able to communicate space and structure accurately to a computer — and that capability survives the AI era.

Second, modeling skill per se (polygon editing, UV, rigging) is something students are already starting to replace with AI; within the 90-hour budget, teaching it deeply has poor cost-benefit.

Third, if you take it to zero, students lose the ability to recover when AI goes off the rails (composition breaks, space stops being coherent).

Note: if there is a meaningful cohort of art-school-background students who want the modeling feel, Option C is better handled by beefing up self-study material rather than burning class time on it. Pointing them at high-quality external material like the Blender Donut Tutorial — Blender Guru (2025) works perfectly well.

5. Risks and caveats

AI video generation services have fluid terms of service, copyright policies, and training-data positions. Plan on reviewing the services used in class every year
Pricing on Comfy Cloud and Runway shifts. Confirm in advance whether the university budget can cover per-student accounts
Some students will misread this as “AI generates everything in one shot.” Make Price’s “labor vs. judgment” frame explicit in session 1 and reframe the practice as “stack judgment on top of the first AI output”
30 sessions / 90 hours is plenty for production-centered practice, but if too much time goes to technical exposition, production time gets squeezed. Push most of the technical exposition to pre-recorded video and self-study material, and keep in-class time concentrated on “production and judgment”
On a 5–10 year horizon, real-time generation, interactive generation, and 3D Gaussian / NeRF intake all come into view. Designing the course skeleton to be tool-agnostic keeps revision overhead low

6. Summary

In one line: 3DCG shrinks as a modeling skill but survives as an AI control surface; what remains for humans is judgment, story, and world. From there, the draft direction (“live-action + AI generation at the center”) is correct, but three adjustments raise its durability against the next 5–10 years of change: (1) do not allocate ratios by tool name, (2) do not take 3DCG to zero — keep three sessions in the second semester as “3D for controlling AI,” and (3) redesign the 30 sessions as a whole as practice in judgment.

180 min × 15 sessions × 2 semesters is a sufficient overall budget to balance technical learning and finished work. The skeleton is a two-stage structure — first semester to walk through the first-pass pipeline for asset creation, second semester to push into character consistency, 3D control, and finishing a piece — with the final week always reserved for critique.

Media Creation Curriculum in the AI Era #2 — A ComfyUI-Based Curriculum