Back to Prototypes

Prototype 02

Experimental 68% mapped

Creative Engine

Pipeline-style engine combining generation, annotation, and evaluation across text, audio, and visual formats. From concept to output using agent frameworks, custom prompt templates, and quality assessment protocols.

Core Pipeline

7 pipeline stages · concept to published output

1

movie-making-pipeline

Orchestrator

Source of truth for AI video production. 219 components, 8 domains. Full 51-clip episode verified. Stages: ANALYZE, VALIDATE, REFINE, EXECUTE, REVIEW.

2

creative-engine

Execute

FastAPI service for VEO3 video generation, stitching, and storage. 13 REST endpoints.

3

content-mcp

Generation

34 MCP tools. Intent to PromptContract objects — traceable, reusable, evaluable prompt specifications.

4

content-pipeline-mcp

Assembly

TTS audio, mixing, video stitching, frame extraction, episode loading.

5

content-analyzer-mcp

Analysis

9 tools. Gemini multimodal review, aesthetic classification, outlier detection (Isolation Forest).

6

evaluation-mcp (Evalla)

Quality

6 tools. Rubric-based scoring, HumilityAuditor for 8-point claim substantiation.

7

production-creator-mcp

Validation

7 tools. Four-asset validator (TEXT/VISUAL/OPERATIONAL/KNOWLEDGE), 30 schemas.

Audio Layer

Voice synthesis, music generation, and distribution

voice-mcp

ElevenLabs TTS with disk cache and voice profiles

the-musicologist-mcp

17 tools. Music analysis frameworks, vocal annotation, platform formatting, copyright safety, Gemini review

MELODIA

Music release engine — create, produce, distribute across platforms

flow-musical-creation

Audio productions — multi-act, multi-song structures with metadata and musicological analyses

Visual Layer

Video generation, animation, and image analysis

creative-engine

VEO3 video generation service

CC Video generation

Video generation and assembly pipeline

Music-Video-Creator

End-to-end music video production

cartoon_maker

Cartoon and animation generation

content-extractor-mcp

Screenshot analysis, UI region detection, OCR text extraction

Prompt & Content Management

Prompts as composable objects, not just text strings

Project-Prompt-as-a-Tool

Dynamic prompt-to-tool conversion via MCP. Sequential chaining, Jinja2 templates, runtime config

quick-prompt-manager

Prompt template management and retrieval

AutomaticContentCreator

Multi-model competitive generation (Claude + GPT-4o + Gemini) with review and winner selection

meta-prompt-to-asset-factory

Framework for "From Meta-Prompt to Asset Factory" methodology

Architecture

Creative Engine Architecture diagram showing Concept/Brief flowing through content-mcp PromptContracts into Text, Audio, and Visual branches, converging at content-pipeline-mcp, with three quality outputs: content-analyzer-mcp, Evalla, and production-creator-mcp

Capability Coverage

Claim Implementation Status
Pipeline-style engine movie-making-pipeline (5-stage E2E verified) Production
Generation (text) content-mcp (34 tools), AutomaticContentCreator Production
Generation (audio) voice-mcp, MELODIA, flow-musical-creation Production
Generation (visual) creative-engine (VEO3), cartoon_maker, Music-Video-Creator Production
Annotation content-analyzer-mcp, the-musicologist-mcp, content-extractor-mcp Production
Evaluation evaluation-mcp (Evalla) rubrics, HumilityAuditor Production
Agent frameworks SPINE orchestration, content-mcp PromptContracts Production
Prompt templates Project-Prompt-as-a-Tool, quick-prompt-manager Production
Quality assessment Evalla, production-creator-mcp 4-asset validator Production
Cross-format integration Audio + visual + text in single automated pass Partial

All pipeline stages exist as production code. The remaining gap is cross-format integration — running audio, visual, and text in a single automated pass from concept to published output. 29 projects compose this prototype.