Download the PHP package netresearch/nr-repurpose without Composer

nr_repurpose — Content Repurpose for TYPO3

Turn a webpage (URL) or PDF into three AI-generated media artifacts — a podcast with one to three persona-driven speakers (with transcript + WebVTT subtitles), a diagram (Schaubild, in three variants), and an Instagram-story carousel — from the TYPO3 backend.

Every AI call goes through netresearch/nr-llm: nr_repurpose contains no provider code. Any LLM, image or TTS provider works — if nr-llm supports it (see Providers, models and prompts).

What it produces

From one source (URL or PDF) the pipeline derives a single faithful ContentBrief (via nr-llm, source language auto-detected) and generates:

• Podcast — a dialogue between one to three speakers: select up to three persona snippets, each contributing a speaker name, a character description for the script and optionally its own TTS voice — or get the classic two-host default. Synthesized turn-by-turn via nr-llm's text-to-speech service, stitched with ffmpeg into one MP3, plus a speaker-tagged transcript and a WebVTT subtitle file whose cue times come from the measured segment durations.
• Schaubild — three variants for comparison: pure HTML (Fluid → headless Chromium → PNG), HTML with an AI-generated background, and a full AI image at the dimensions the selected layout snippet defines. Branded NR or neutral theme.
• Story — a multi-slide carousel (9:16 slides): a cover hook, one slide per key point (at most four) and an outro with the source attribution — up to six slides, one artifact per slide. A single optional AI background is shared by all slides — generated at the layout-selected dimensions and scaled to cover the design canvas so the layout is never distorted.

Each artifact type can be selected per run. Long-running generation runs asynchronously via Symfony Messenger (doctrine transport).

Editorial steering and transparency

• Prompt snippets — the job form offers audience, tone of voice, persona, layout and style selectors, populated from nr-llm's prompt-snippet library (each option shows its description). A layout snippet's imageSize metadata drives the AI-image dimensions per channel (skyscraper, wide, square, …). A fresh installation has no snippets, so install the Content Repurpose Starter pack to fill the five selectors — from nr-llm's Use Case Packs module, or vendor/bin/typo3 nrllm:usecasepack:install content-repurpose-starter.
• Live progress — while a job runs, the detail view shows fine-grained per-step progress and refreshes itself.
• Prompt transparency — every generated artifact records its complete creation parameters: the exact system, user and image prompts, the models, image sizes and voices used. They are shown in the job detail view.

Providers, models and prompts

nr_repurpose never picks a provider itself — it names nr-llm Configuration records (use cases) and lets nr-llm resolve the model, provider, API key, system prompt and cost tracking:

You do not create these records by hand: nr_repurpose declares them as configuration presets (nr-llm ADR-056). Open nr-llm's Configurations backend module — the three nr_repurpose_* records appear as pending presets with their required capabilities, and a single click imports each as a criteria-mode configuration that resolves against the models you have. Mark the imported nr_repurpose_text record as the instance default for the analysis/copy calls.

System prompts (e.g. the image-style preamble) are maintained on the Configuration records; per-model and per-configuration usage and cost show up in nr-llm's analytics module. API keys belong to nr-llm and are referenced by identifier (e.g. nr_repurpose_openai) — no plaintext key ever lives in extension configuration (nr-llm ADR-030).

Honest limits today: text generation is fully provider-agnostic; image and speech go through nr-llm's specialized services, which currently cover OpenAI (images, TTS) and fal.ai (images). The extension-side seam is in place — ImageGeneratorInterface / SpeechSynthesizerInterface with a DI alias in Configuration/Services.yaml — so a fal.ai image backend is a small adapter class away, and additional providers become available as nr-llm grows them.

Full control — no black box

nr_repurpose is not a SaaS pipeline you feed content into and hope for the best. It runs entirely inside your TYPO3 instance, and through nr-llm every aspect of the AI usage stays under the operator's control:

• Provider sovereignty — decide per use case which provider serves it: a US cloud, an EU provider (e.g. Mistral), or fully self-hosted models via Ollama, where content never leaves your infrastructure. Switching is a backend record edit, not a deployment.
• Costs — per-user budgets are enforced by nr-llm's middleware; every call is metered and attributed per model and per configuration in nr-llm's analytics module; image and speech calls are additionally pre-gated against the budget with a planned cost before any money is spent.
• Prompts — system prompts are maintained centrally on Configuration records, editorial steering on reviewable prompt snippets, and every artifact stores the exact prompts, models, sizes and voices that produced it — reproducible and reviewable after the fact.
• Auditing — nr-llm keeps the API keys encrypted and routes every outbound AI call through an audited client: a who/what/when trail exists for each one.
• Permissions — backend group permissions gate which editors may spend on audio (generate_audio) and AI imagery (generate_vision).

Requirements

• TYPO3 v14.3 LTS, PHP 8.3+
• nr-llm ^0.30 (installed automatically via Composer; it owns the provider credentials)
• An API key for at least one nr-llm-supported provider. The tested default stack uses a single OpenAI key for everything (analysis, TTS, images).
• ffmpeg, poppler-utils and chromium (+ Node.js for the renderer) on the host that runs the worker — baked into the DDEV web image.

Local development (DDEV)

Prerequisites: Docker + DDEV.

The bundled dev wiring uses OpenAI: ddev install seeds the key into nr-vault under nr_repurpose_openai and wires nr-llm's provider, so no further configuration is required for a dev instance.

Backend: https://nr-repurpose.ddev.site/typo3/ — user admin, password Demo1234!. Open Web › Repurpose, choose New job, paste a URL, pick the artifacts, theme and prompt snippets, and submit. The list shows the job progress; the detail view shows live per-step progress, plays the podcast (with subtitles + transcript), shows/downloads every image, and lists the exact prompts and models behind each artifact.

CLI

Run the full pipeline for a job synchronously (useful for ops / debugging without the async worker):

Tests

Always run via the Docker-isolated runner (TYPO3 core-testing images, default PHP 8.5) — never inside ddev:

Architecture

See the rendered documentation under Documentation/ (Introduction, Installation, Configuration, Usage, Architecture, and the Architecture Decision Records). Pipeline: ingest (web/PDF) → analyze (one ContentBrief via nr-llm) → generate (podcast / schaubild×3 / story×N slides) → store in the TYPO3 File Abstraction Layer (FAL).