PoRedoImage — AI-Powered Image Studio

Product Requirements Document (PRD) · v2.0 · April 2026

---

...

Product Vision

PoRedoImage is a cloud-native AI image studio that transforms ordinary photos into artistic masterpieces, memes, and stylistic variations in seconds. By chaining Azure Computer Vision, Azure OpenAI gpt-5.4-nano (text and vision), and Google gemini-2.5-flash-image behind a clean Blazor Web App, PoRedoImage makes professional-grade AI image manipulation accessible to anyone — no prompt engineering required.

The core user promise: upload a photo, choose a style, get a gallery-worthy result in under 10 seconds.

---

Product Requirements

Goals

1. Instant AI Transformation — Users can upload any JPEG/PNG and receive an AI-regenerated image or a captioned meme within one interaction.
2. Bulk Art Studio — Power users can generate 10 distinct art-style variations of a photo in a single click, with results streaming live as each slot completes.
3. Personal Gallery — Every result can be saved to a persistent per-user gallery backed by Azure Table Storage, accessible across sessions.
4. Zero-friction Auth — Development environment uses a one-click cookie login; production uses Microsoft Entra ID OIDC with no additional friction for M365 users.
5. Observable & Reliable — Every AI call is traced via OpenTelemetry and logged via Serilog to Application Insights; a /health endpoint verifies all dependencies at runtime.

Non-Goals (v1)

• Native mobile app (responsive web only)
• Video processing
• Real-time collaborative editing
• Custom model fine-tuning UI

User Personas

Persona	Core Need	Primary Flow
Creative — social media creator	Unique art variations for posts	Bulk Generate × 10 styles
Casual — personal user	Fun meme from a photo	Meme Generation mode
Developer — API consumer	Integration testing + diagnostics	/diag + /scalar/v1 + /health

Key Metrics (Success Criteria)

Metric	Target
End-to-end image regeneration latency	< 10 s p95
Bulk generate (10 variations) wall-clock	< 45 s p95
CI test coverage gate	≥ 80% (opencover)
Production deployment success rate	≥ 99% (OIDC zero-secret deploy)
/health uptime SLA	99.5%

---

Architecture Overview

flowchart LR
    User["👤 User"] -->|"HTTPS"| App["Blazor Web App\nAzure App Service"]
    App -->|"AI Calls"| AI["CV · OpenAI · Gemini"]
    App -->|"Persist"| Data["Table Storage\n+ Key Vault"]
    App -->|"Telemetry"| Ops["Application Insights"]
    CI["GitHub Actions\nOIDC"] -->|"Deploy"| App

    style User fill:#4a9eff,stroke:#2a7fd4,color:#fff
    style App fill:#512bd4,stroke:#3a1fa8,color:#fff
    style AI fill:#10a37f,stroke:#0a7a5f,color:#fff
    style Data fill:#f2c811,stroke:#c9a000,color:#000
    style Ops fill:#0078d4,stroke:#005fa3,color:#fff
    style CI fill:#238636,stroke:#196228,color:#fff

Loading

Live: https://poredoimage-web.azurewebsites.net | API Docs: /scalar/v1 | Health: /health

---

Key Features

Feature	Description
Image Analysis	Computer Vision → gpt-5.4-nano enhancement → tags + confidence
Image Regeneration	Gemini gemini-2.5-flash-image with reference bytes
Meme Generation	SkiaSharp text overlay on analysed image
Bulk Generate	10 art-style variations via parallel Gemini calls, streamed live
Auth	Dev: /dev-login cookie · Prod: Microsoft Entra ID OIDC
Diagnostics	/diag masked config · /health · /scalar/v1 API docs

---

Tech Stack

Layer	Technology
Framework	.NET 10 Blazor Web App (global Interactive WebAssembly, no prerender) behind an ASP.NET Core BFF
AI — Vision	Azure Computer Vision cv-poshared-eastus
AI — Language + Vision	Azure OpenAI gpt-5.4-nano — one deployment serves reasoning and image-to-text
AI — Image Gen	Google Gemini gemini-2.5-flash-image
AI — Music	Google Lyria lyria-3-clip-preview (Rap Roast)
Storage	Azure Table Storage stporedoimage26
Secrets	Azure Key Vault kv-poshared (Access Policy + 30 min rotation)
Observability	OpenTelemetry + Serilog → Application Insights
Infrastructure	Azure Bicep + GitHub Actions OIDC
Testing	xUnit · Testcontainers · C# Playwright (Unit · Integration · E2EAPI · E2EUI) — not run in CI

---

Documentation

Architecture

Live Mermaid diagrams live alongside their source in docs/. See docs/README.md for the full index of architecture, journey, state, data, and UI diagrams.

---

Getting Started

Prerequisites

• .NET 10 SDK
• Azure subscription with Computer Vision + OpenAI resources

1. Clone and restore

git clone https://github.com/punkouter26/PoRedoImage.git
cd PoRedoImage
dotnet restore PoRedoImage.slnx

2. Configure secrets (local)

There is no local secret store to populate — dotnet user-secrets is deliberately not used (the project has no UserSecretsId). Local runs read the same Azure Key Vault the deployed app does, through DefaultAzureCredential. Sign in once and the host picks the secrets up on start:

az login          # the signed-in identity needs "Key Vault Secrets User" on kv-poshared
dotnet run --project src/PoRedoImage.Web

AddPoRedoImageKeyVault loads KeyVault:Uri (https://kv-poshared.vault.azure.net/) and StartupSecretValidator fails the host fast, naming any secret it could not resolve. To work without Key Vault access — no AI calls, no storage — run against the mocks instead:

Mocks__UseMockAi=true Storage__ConnectionString="" dotnet run --project src/PoRedoImage.Web

3. Run

dotnet run --project src/PoRedoImage.Web
# → http://localhost:4000  |  https://localhost:4001
# Dev login: http://localhost:4000/dev-login?email=you@example.com

4. Test

dotnet test PoRedoImage.slnx                                    # Unit + Integration
dotnet test tests/PoRedoImage.Tests.E2E                          # E2E (C# Playwright + HTTP smoke)

---

API Endpoints

Method	Path	Description
GET	/health	Full health check (JSON)
GET	/alive	Liveness probe
GET	/diag	Masked config diagnostics
POST	/api/images/analyze	Analyze + process image
GET	/api/bulk-generate/prompts	Load saved art prompts
POST	/api/bulk-generate/prompts	Save art prompts
GET	/scalar/v1	Interactive API docs

---

Project Structure

src/PoRedoImage.Web/        # API/BFF host
  Features/
    Auth/             # OIDC + dev login cookie handler, /auth + /api auth
    BulkGenerate/     # Imagen3Service, parallel generation, prompt storage endpoints
    Diagnostics/      # /api/diag endpoint, middleware
    ImageAnalysis/    # ComputerVisionService, OpenAIService, MemeGeneratorService
  Components/         # App.razor host document + _Imports (renders the Client's <Routes> as WASM)
  Configuration/      # KeyVaultSecretNameMapping
src/PoRedoImage.Client/     # Blazor WASM SPA
  Routes.razor        # Router (global InteractiveWebAssembly)
  Pages/ Layout/ Shared/ Models/   # all interactive UI + ImageSessionService
  wwwroot/            # static assets (the only wwwroot in the solution)
tests/
  PoRedoImage.Tests.Unit/            # xUnit pure logic
  PoRedoImage.Tests.Integration/     # xUnit + WebApplicationFactory + Testcontainers
  PoRedoImage.Tests.E2EAPI/          # pure HTTP API E2E (xUnit, self-skip if no live instance)
  PoRedoImage.Tests.E2EUI/           # C# Playwright UI E2E (self-skip if no live instance)
infra/
  main.bicep          # App Service + Storage provisioning
docs/                 # All .mmd diagrams + screenshots

---

Key Vault Secrets (Production)

All secrets load from kv-poshared via AZURE_KEY_VAULT_ENDPOINT app setting.

Key Vault Secret	Config Key
PoRedoImage-ComputerVision-ApiKey	ComputerVision:ApiKey
PoRedoImage-ComputerVision-Endpoint	ComputerVision:Endpoint
PoRedoImage-OpenAI-ApiKey	OpenAI:Key
PoRedoImage-OpenAI-Endpoint	OpenAI:Endpoint
PoRedoImage-StorageConnectionString	Storage:ConnectionString
PoRedoImage-Google-ApiKey	Google:ApiKey
PoRedoImage-Google-Imagen3Model	Google:Imagen3Model
PoRedoImage-ApplicationInsights-ConnectionString	ApplicationInsights:ConnectionString

OpenAI:ChatCompletionsDeployment is deliberately NOT a Key Vault secret. The deployment name is not sensitive, and a stale Key Vault copy (gpt-4.1-nano) once shadowed the live value and caused 404 DeploymentNotFound. The single source of truth is config: appsettings.json for local/dev/test, overridden in production by the literal OpenAI__ChatCompletionsDeployment app setting in infra/main.bicep. Do not re-add a PoRedoImage-OpenAI-DeploymentName secret — see the note in KeyVaultSecretNameMapping.cs.

---

Dev Guidelines

• Vertical Slice Architecture — all feature files in Features/{Name}/
• Minimal APIs — no MVC controllers; use MapGroup + static handler methods
• Nullable + warnings as errors enforced via Directory.Build.props
• Prefix: PoRedoImage for all namespaces and Azure resources