Inspect and generate media with ak:ai-multimodal

Use ak:ai-multimodal to turn image, audio, video, or document inputs into reviewable analysis, transcripts, structured extraction, converted documents, or generated media. The Skill orchestrates the exact-pinned @mrgoonie/multix@0.2.0 CLI; provider catalogs, not the Skill, determine current models, formats, availability, limits, and pricing.

Choose ak:ai-multimodal for media work

Use ak:ai-multimodal when

• You need OCR, visual analysis, transcription, structured extraction, or document-to-Markdown conversion.
• You want to generate or transform an image, video, speech, or music asset through a configured provider.
• You can supply the media, desired output, review criteria, and permission to send the input to the selected provider.

Choose another workflow when

• Text-only inspection is sufficient and no provider upload is needed.
• You need asset organization rather than media processing.
• You cannot permit the source media or prompt to leave the current environment.
• You need a guaranteed model, price, duration, format, or quality level. Those are live provider properties and must be verified before the run.

Prepare Node, tools, and provider access

Before you start:

• Complete Onboarding, install Marketing Kit, and use Node.js 20 or later.
• Pre-warm the exact pin with npx -y -p @mrgoonie/multix@0.2.0 multix --version when later execution may be network-restricted. The first npx run can fetch the package.
• Run npx -y -p @mrgoonie/multix@0.2.0 multix check --verbose.
• Provide only the key required by the chosen path: GEMINI_API_KEY, optional OPENROUTER_API_KEY, or optional MINIMAX_API_KEY. Keep real values out of prompts, files under version control, and final reports.
• Check the selected provider's current catalog and pricing, then record an explicit compatible model when reproducibility matters.
• Confirm FFmpeg or other local media support reported by multix check.

See Runtime adapters before assuming equivalent tool or approval behavior.

Run a reviewable analysis

/ak:ai-multimodal Analyze ./research/interview.mp4, create a timestamped Markdown transcript and a separate evidence summary, redact personal identifiers, and do not generate or publish media

/ak:ai-multimodal Analyze ./research/interview.mp4, create a timestamped Markdown transcript and a separate evidence summary, redact personal identifiers, and do not generate or publish media

$ak:ai-multimodal Analyze ./research/interview.mp4, create a timestamped Markdown transcript and a separate evidence summary, redact personal identifiers, and do not generate or publish media

For generation, name the modality, brief, output path, review criteria, chosen provider, and model already verified against the live catalog. Do not ask the Skill to infer “latest,” “default,” or “recommended.”

Follow the media stages

1. Define the contract. Identify input paths, modality, expected output, privacy constraints, provider, and acceptance criteria.
2. Check the environment. Verify Node, the pinned CLI, provider key, local media tooling, access, and current provider capability.
3. Choose the operation. Use analysis, transcription, extraction, document conversion, generation, or media optimization as appropriate.
4. Process with explicit controls. Use the pinned command syntax and an explicit verified model for generation. Split oversized media only after checking current input and output limits.
5. Inspect the artifacts. Check completeness, timestamps or structure, factual support, output format, and provider errors.
6. Report evidence. Name every output, provider path, verified model when applicable, command result, redaction, and unresolved limitation.

Keep uploads, spend, and publication separate

Do not retry a provider error blindly. Preserve its useful detail, redact credentials, and resolve authentication, billing, quota, access, or request parameters first. Review generated media for rights, privacy, brand fit, factual claims, and unsafe content before any external use.

Verify the outputs

A complete run should provide:

• Exact input and output paths without exposing secrets.
• The operation, provider route, and explicit model used for generation.
• multix check or equivalent environment evidence.
• A transcript, analysis, structured file, converted document, or media artifact that meets the stated format.
• Missing segments, failed chunks, redactions, and current provider limits.
• A clear statement that publication, distribution, or downstream upload did not occur unless separately approved.

Troubleshoot and understand limits

The exact CLI pin changes only when the Kit is refreshed. Provider behavior can change independently, and the Skill makes no quality, availability, model, price, or performance guarantee. Use Marketing Kit lifecycle when the packaged backend needs an approved refresh.