Vertex AI Creative Agent & Supabase Pipeline

An autonomous end-to-end B2B marketing engine: OpenAI GPT-4o synthesizes emotionally gripping ad prompts, Google Cloud Vertex AI Imagen 3 generates photorealistic visuals, and Supabase Storage + PostgreSQL persists assets with a standardized contract to power tomorrow's connected distribution puzzle.

💡 Challenge Clue: Think pipeline, prompt > generation > storage

Connected Puzzle Part 1 of 2: Today’s output sets status: 'ready_for_distribution' in Supabase, providing the upstream asset feed for tomorrow’s social distribution puzzle (#38).

Bridge Contract: Active

Autonomous Workflow Canvas

Status: Ready

⏱

Schedule Trigger1

scheduleTrigger

IDLE

🤖

AI Agent3

langchain.agent

IDLE

🧠

OpenAI Model3

gpt-4o

💾

Simple Memory1

bufferWindow

{}

Code6

parsePrompt

IDLE

✎

SET1

manualEnv

IDLE

🔑

JWT1

signRS256

IDLE

🌐

GET TOKEN1

oauth2.token

IDLE

{}

Code5

prepVertexReq

IDLE

🎨

Generate Image1

vertex.imagen3

IDLE

{}

Code7

prepSupabase

IDLE

⚡

HTTP Request2

storage.upload

IDLE

⚡

Create a row

supabase.insert

IDLE

Select any node on canvas to inspect live telemetry

Interactive Inspector

// Click on any node above to view its structured JSON payload and input/output contracts.

Active Asset Output

STATUS: STANDBY

"STILL DOING IT ALL BY HAND?"

Cinematic, hyperrealistic photograph of a fatigued male professional at a dark office desk rubbing his forehead. Foreground office desk telephone with glowing red-orange LED displaying "17 MISSED CALLS".

Storage: ad-creatives/creative-still-doing-it.png

AI Prompt Synthesis Engine

Model: GPT-4o

Marketing Pain-Point Preset

Campaign Headline Overlay

Central Emotional Subject & Narrative Prop

Camera & Lighting Directive

Synthesized Prompt Directive

Full Prompt (Positive)

Negative Prompt

cartoon, illustration, 3d render, distorted anatomy, extra fingers, blurry text, low resolution, watermark

Aspect Ratio: 1:1 (Square Ad)

Sample Count: 1

Format: PNG

Google Cloud Vertex AI Configuration

OAuth2 / Service Account

Google Cloud Project ID

Region & Vertex Endpoint

Foundation Model

JWT RS256 Assertion Status

eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJ2ZXJ0ZXgtc2FAcHJvamVjdC5pYW0uZ3NlcnZpY2VhY2NvdW50LmNvbSIsImF1ZCI6Imh0dHBzOi8vb2F1dGgyLmdvb2dsZWFwaXMuY29tL3Rva2VuIiwic2NvcGUiOiJodHRwczovL3d3dy5nb29nbGVhcGlzLmNvbS9hdXRoL2Nsb3VkLXBsYXRmb3JtIiwiZXhwIjoxNzI2MjMzMDAwfQ...

Vertex AI Response Inspector

Latency: 1,420ms

Predicted Base64 Image Payload

{ "predictions": [ { "mimeType": "image/png", "bytesBase64Encoded": "iVBORw0KGgoAAAANSUhEUgAABAAAAAQACAIAAADHA7gRAAA...", "aspectRatio": "1:1", "safetyAttributes": { "blocked": false, "categories": [] } } ] }

Decoded buffer size: 2,841,920 bytes (2.7 MB) · Direct pass to Supabase Storage buffer stream.

Supabase PostgreSQL: creative_generations Table

Bucket: generated-images

ID	Headline	Model	Storage Path	Status (Contract)	Created At	Actions

Puzzle #38 Consumer Query: SELECT * FROM creative_generations WHERE status = 'ready_for_distribution' ORDER BY created_at ASC LIMIT 1;

Challenge Benchmark Creative Dissection

Interactive Hotspots

Prompt Engineering Secrets

Vertex AI Imagen 3

1. In-Image Typography Rendering: Imagen 3 excels at rendering legible text when enclosed in exact quotation marks with explicit placement directives (e.g. across the top in bold condensed white sans-serif letters reading "STILL DOING IT ALL BY HAND?").

2. Emotional Grounding: Rather than generic adjectives ("tired"), prompting specific micro-expressions ("head resting in palm, furrowed brow, loosened navy tie, cognitive burnout") yields authentic, conversion-focused B2B creative imagery.

3. The Narrative Prop Hook: Adding a tangible quantified pain-point (the desk phone illuminated with "17 MISSED CALLS") immediately anchors the viewer’s attention and communicates loss aversion.

4. Camera Optics: Simulating an 85mm portrait lens at f/1.8 with shallow depth of field isolates the overwhelmed professional while maintaining atmospheric context in the background.

8-Point Automated Diagnostic Suite