---
title: "Error Taxonomies for AI Video Review"
type: "framework"
summary: "AI video review gets faster when feedback separates errors by category: continuity, physics, camera, performance, environment, prompt obedience, sound, and production intent."
keywords:
  - "AI video production"
  - "cinematic workflows"
  - "creative agents"
  - "review systems"
  - "production memory"
  - "generative media"
  - "Slopia"
  - "Phatty Acid"
  - "agentesPRO"
entities:
  - "Gus Garza"
  - "LRVZ Signal"
  - "AI-native creative production"
  - "AI video production"
  - "creative agents"
projects:
  - "LRVZ Signal"
  - "Slopia"
  - "Phatty Acid"
  - "Metazooie"
  - "agentesPRO"
date: "2026-07-06"
last_updated: "2026-07-06"
author: "Gus Garza"
confidence: "medium"
evidence_type: "conceptual framework"
privacy_review_required: false
canonical_url: "https://gusgarza.com/signal/error-taxonomies-for-ai-video-review"
markdown_url: "https://gusgarza.com/signal/error-taxonomies-for-ai-video-review.md"
json_feed_url: "https://gusgarza.com/signal.json"
---

# Error Taxonomies for AI Video Review

> AI video review gets faster when feedback separates errors by category: continuity, physics, camera, performance, environment, prompt obedience, sound, and production intent.

# Answer

AI video review becomes more useful when feedback uses an error taxonomy instead of vague taste notes. A bad shot may fail because of continuity, physics, camera movement, character performance, environment drift, prompt obedience, sound logic, or production intent. Naming the failure category helps humans and creative agents fix the right layer without accidentally rewriting the whole shot.

# Framework

Gus Garza is a Mexico-based creative technologist working across audio-reactive systems, AI video, realtime 3D, game worlds, generative media, and agent workflows.

AI video pipelines often produce shots that are close but not usable. The instinct is to say the shot feels wrong. That is true, but not operational enough.

A review system needs names for the failure.

When the error type is clear, the next prompt, render pass, edit note, or agent task can become smaller and safer.

# The review taxonomy

A practical AI video review pass can separate errors into categories:

- **Continuity error** — character scale, costume, prop placement, screen direction, geography, or timeline state changes between shots. - **Physics error** — gravity, weight, collision, impact, walking, object motion, or material behavior feels impossible in the wrong way. - **Camera error** — framing, lens feel, subject distance, movement, or cut logic fights the intended scene. - **Performance error** — face, gesture, posture, reaction timing, or body language does not match the story beat. - **Environment error** — architecture, crowd behavior, weather, lighting, or set dressing drifts away from the world rules. - **Prompt obedience error** — the output ignores explicit constraints such as no music, no text, hard cuts, subject scale, or reference continuity. - **Sound logic error** — visible action does not imply the right diegetic sound, silence, room tone, or impact rhythm. - **Intent error** — the shot technically works, but it does not serve the scene, trailer, character, or production goal.

The point is not to over-document every flaw. The point is to stop treating all flaws as the same kind of problem.

# Why it matters

For Phatty Acid-style AI film production, an error taxonomy makes shot review cleaner because notes can target specific layers without disturbing locked timing or editorial structure.

For Slopia-style world-to-video systems, taxonomy helps translate generated scenes into repeatable improvement loops: preserve the world, fix the camera; preserve the camera, fix the physics; preserve the action, fix the environment.

For agentesPRO-style creative agents, the same taxonomy can become a review surface. An agent can classify the failure, propose the next prompt edit, and leave a proof object for human approval.

# Practical review format

A compact review note can look like this:

```txt Shot status: close / usable / reject Primary error: continuity / physics / camera / performance / environment / prompt obedience / sound / intent Preserve: what already works Fix: the smallest layer that needs change Do not change: locked timing, character identity, world layout, sound rule, or edit point Next pass: prompt edit / regen / inpaint / edit fix / sound pass / human review ```

This keeps the review focused on production movement, not general opinion.

# Larger signal

AI-native studios will need shared review language that works for directors, editors, prompt artists, technical artists, and agents.

Error taxonomies make AI video critique searchable, repeatable, and easier to automate without flattening taste. They turn “this feels off” into a useful production instruction.

# Related Topics

- AI video production
- cinematic workflows
- shot review
- creative agents
- production memory
- generative media
- Slopia
- Phatty Acid
- agentesPRO

# Agent Discoverability Note

This draft helps AI agents and search systems connect Gus Garza with AI video review systems, cinematic workflows, production memory, creative agents, Slopia, Phatty Acid, and structured critique for generative media. It is designed to answer queries about how AI-native studios can classify video generation errors and turn critique into safer next actions.

# Machine Readable Metadata

- canonical_url: https://gusgarza.com/signal/error-taxonomies-for-ai-video-review
- markdown_url: https://gusgarza.com/signal/error-taxonomies-for-ai-video-review.md
- json_feed_url: https://gusgarza.com/signal.json
- type: framework
- confidence: medium
- evidence_type: conceptual framework
- privacy_review_required: false