pictovap 0.2.1

Open-source visual finishing engine for content publishers.

Pictovap

Pictovap is an open-source visual finishing engine for content publishers. It turns a finished article into a visually complete, rights-aware, publish-ready CMS page — and shows its work at every step.

The Problem

Every publisher runs the same manual routine before an article can go live: find images that actually fit the section they're going into, check whether the license permits the intended use, resize and convert them to the site's format, write alt text and a caption for each one, and place them at the right point in the CMS. None of this is hard, individually. It is, however, repetitive, easy to get wrong under deadline pressure, and it scales linearly with how much a publisher writes. Skipped alt text quietly erodes accessibility and SEO. Untracked image provenance is a license or attribution problem waiting to surface later, when it's expensive to fix. None of this shows up in a style guide — it shows up as an hour of an editor's afternoon, per article, forever.

Point solutions exist for pieces of this: stock photo plugins, DAM systems, generic AI image generators. What's largely missing is the connective layer — something that reads what the article actually needs, evaluates candidates against that need with a visible, auditable reason for every accept and reject, and hands a CMS a placement plan it can execute without a human re-deriving the same context from scratch.

The Solution

Pictovap is that connective layer, built as an open, adapter-based pipeline rather than a closed SaaS product:

Article Input → Visual Brief → Candidate Images → Fit Score
              → Provenance Pack → CMS Placement → Editor Report

• Visual Brief — a structured read of what imagery the article actually needs, derived from its heading structure and content, not from a generic "insert 3 images" rule.
• Fit Score — every candidate is scored against the brief with a transparent, deterministic reason attached (selected, rejected, or needs_review, plus why). No black-box relevance number.
• Provenance Pack — a persistent audit trail for every selected image: source, license status, attribution, a content hash, and the exact processing actions applied. This is what makes "where did this image come from and are we allowed to use it" answerable six months later.
• CMS Placement — a CMS-agnostic plan describing where and how each image should be placed, independent of whether the destination is WordPress, Ghost, Strapi, or something a contributor writes an adapter for tomorrow.
• Editor Report — a human-readable Markdown review surface, so an editor signs off on a report, not raw JSON, before anything reaches production.

What makes this radical isn't any single stage — it's that the whole pipeline is a public, inspectable contract instead of a hosted black box, and that it is honest about its own workings by default: the credential-free demo below runs the entire pipeline end-to-end, with zero API keys and zero network calls, so you can read exactly what it does before you ever hand it a real site.

Pictovap is not a stock photo search tool, a DAM, a generic AI image generator, or a WordPress-only plugin. It has no graphical interface yet — it is a CLI-first, adapter-based core, with the editor report as the intended human review surface and CMS adapters as the machine-facing execution layer.

Quickstart

git clone https://github.com/yoldaolmak/Pictovap.git
cd Pictovap
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pip install -e .

Run the credential-free demo:

pictovap demo

  Brief:      4 slots from 3 sections
  Evaluated:  5 candidates
  Selected:   3 images
  Rejected:   4 candidates
  Placements: 3 instructions

No .env file, no API keys, no network calls — every candidate and score above comes from deterministic mock data, on purpose. This is the demo's guarantee, not just its default state.

Try Your Own Article

pictovap plan \
  --article path/to/your/article.md \
  --profile examples/profiles/sample-publisher.yaml \
  --output my-plan.json \
  --report my-report.md

my-plan.json is the canonical, machine-readable artifact for adapters and automation. my-report.md is the same plan, rendered for a human editor to review before anything gets published.

Adapters

Pictovap connects to the outside world only through adapters — the core pipeline has no hardcoded dependency on any specific image provider or CMS.

Image sources: local folder, Unsplash, DepositPhotos, Openverse (no key required), and Pexels are implemented. Pixabay and Wikimedia Commons are open contribution opportunities — see Good First Issues. See Image Source Adapters.

CMS placement: WordPress (production-tested), Ghost and Strapi (reference implementations, real but with documented limitations). See CMS Adapters.

Every adapter degrades gracefully when unconfigured — a missing API key produces an empty result, not a crash, so a partially configured profile still runs. Writing a new adapter means implementing one method (search_candidates or place) against a documented Protocol; see the Adapter Overview.

Multi-Language by Design

Pictovap's own code, comments, and documentation are English. What it generates — alt text, captions, titles — is not tied to any one language. Article language is detected automatically (or set explicitly via the publisher profile), and generated metadata follows it: a Turkish article gets Turkish alt text, a French one gets French, and so on. This isn't a localization afterthought bolted onto an English-only tool; it's a parameter of the pipeline from the start.

Current Status and Limitations

Pictovap is early open-source infrastructure. It has a genuinely credential-free local demo, documented core primitives, a real test suite, and a working adapter model — it does not yet claim broad ecosystem adoption or a large external contributor base.

Specifically:

• Only the WordPress CMS adapter is production-hardened; Ghost and Strapi are real, tested reference implementations with documented gaps (see CMS Adapters).
• The credential-free demo always uses deterministic mock candidates by design, regardless of what's configured in .env — pictovap plan is where real, credentialed sources are used.
• Deterministic structural extraction (the rule-based, non-AI language and section detection) is reliable for English and Turkish; other languages are untested.

Compatibility Note

Product name: Pictovap. The Python package, import name, and console-script entry point may remain pictova for backward compatibility — see Brand & Naming.

Contributing

See the Documentation Portal for architecture, concepts, and adapter-writing guides, and DEVELOPER.md for the contribution workflow. Pull requests that add a new image source or CMS adapter, improve test coverage, or fix documentation are all welcome.

License

MIT — see LICENSE.