Media Workflow

Turn a short request into a complete, reviewable media job. The AI reads your website or item list, plans the assets, produces or prepares them, checks every result, and saves the evidence in your Git repository.

Private preview · 0.1.0. This repository is a private GitHub template. It has not been published to a plugin directory. No public redistribution licence has been selected. Image providers and source assets keep their own terms.

Start with a prompt

Paste this into an AI assistant that can read files, browse and generate images:

Use Media Workflow from this repository. I need photorealistic images for my website https://example.de. Read its public pages and catalogue, work out which images are needed, and create a consistent set. Use my private Git repository <owner/site-repository> for the work. Save the plan, exact prompts, originals, reviews and optimized delivery images. Keep the work private and prepare a reviewed branch; do not publish the website.

example.de is an example input, not a site this toolkit has crawled. Replace the URL and repository. The assistant fills out the working files; you do not need to author JSON. Copy the prompt template.

For existing products:

Use Media Workflow. Here is my product list, including brand and pack size. Find the exact official product pictures and prepare square WebP files and smaller thumbnails for my private site repository. Preserve packaging text, logos and quantities. Use reference-based AI edits only where needed for the requested presentation, and keep the source beside each edited result. Flag uncertain matches and missing images instead of guessing.

The assistant can convert a source image, edit it with a supplied reference, or generate an illustration. It records which happened. A generated photorealistic image is not a camera photograph or an official manufacturer asset.

What the AI produces

1. Brief and inventory: pages/items examined, exact subjects, placements, source links, missing facts, reuse decisions and image count.
2. A workflow: available tools, art direction, sizes/formats, assignments, budget/retry bounds and acceptance criteria.
3. Durable job records: exact prompts, source references, original files, attempts, reviews, corrections and hashes.
4. Delivery: approved assets, thumbnails/variants, page mapping, alt text and validation results. Website publication only when included in your request.

Interruptions are resumable. The AI reads saved attempts and approved work before generating more. A valid checksum never counts as visual approval.

Three ways to use it

This package supplies instructions, templates and a small Python helper. The host must provide website access, media generation/editing and visual/audio inspection. It includes no API key, image model, crawler, paid service subscription or automatic deployment. If generation is unavailable, the AI can prepare the inventory and prompts and report the missing capability. It must not claim that files were generated. Audio/video follow the same record process only when the host has suitable generation and review tools; they are not bundled features.

Files and checks

• General workflow
• Exact product images
• Record format and helper commands
• Prompt template
• Product list template
• Security policy

Python 3.11+ runs the optional helper and tests with no third-party dependencies:

python skills/media-workflow/scripts/media_workflow.py --help
python -m unittest discover -s tests -v
python security/check_dependencies.py
python tools/build_package.py

For checks against a finished job, use verify <project-dir> --complete. Add --require-originals when verifying the original archive too. The normal Git handoff can omit that private archive, but it must say so. See the record guide.

Repository protection requires a pull request, the secret/dependency boundary check, and application tests on Linux and Windows. No public service is deployed by this toolkit. Validation and limitations.