SILELO Media Studio

Scope

Use this skill for media-tool features. Keep it alongside—not instead of—silelo-ai-studio-workflow. This package does not promise that a provider exists, that a long video is rendered, or that a large batch has completed.

Source-linked Web Search

Call an approved server-side search source and normalize results to { title, url, snippet }. Allow only http/https URLs, sanitize text, preserve the source domain, and render outbound links with target="_blank" and rel="noopener noreferrer". Do not fabricate results or citations.

Image Batches

Validate prompt and count, then apply a provider-specific upper bound per run. Return and display both requested and completed. For large workloads such as 100 images, create durable queued jobs with progress, cancellation, retries, and persisted status; do not fire all requests in parallel or describe a queue as completed before artifacts exist.

Browser-side ONNX Vision

Verify model input shape, preprocessing, output labels, and class limitations from the actual model. Serve model and runtime assets from deployment-safe URLs, configure compatible WASM paths, and state whether images stay local or are uploaded. Test with a permitted sample image and report runtime failures honestly.

Video Projects

Treat long-form video as plan → render clips → review → merge. A scene manifest may contain title, duration, scene index, start time, visual prompt, and narration, but it is not a rendered video. Require a connected video provider before offering render or merge controls. Report the maximum verified clip/project size rather than inventing an hour-long output.

Proof Before Delivery

Test web search URL parsing, image batch bounds, ONNX model loading where feasible, video scene count/timeline generation, and success/error notifications. Verify the relevant tool in the browser and state the precise capability level: plan-only, queued, provider-connected, or rendered.