topaz-video-enhancement

What it does

The skill guides automated video enhancement using Topaz Labs models. It covers selecting the appropriate model family (Proteus, Starlight, Astra, Nyx, frame interpolation, and various utilities), setting a parameter surface for tuning (video_type, field_order, auto, focus_fix_level, etc.), and driving the Topaz Platform REST API in an asynchronous multi-step workflow (create, accept, upload, complete-upload, poll status) to process video (and secondary stills via the Image API). It includes guidance on when to apply enhancement (upscaling, denoising, deinterlacing, stabilization, frame interpolation, SDR→HDR, colorization) and when not to apply it (no generation, no non-Topaz tools, no purely editorial tasks). It also provides model-selection heuristics for common footage problems and cautions about artifact risks (over-smoothing, temporal shimmer, face hallucination).

How it works

• Choose the model family based on dominant problem, then the variant by quality, speed, and source condition (Proteus for general upscaling; Iris for faces; Artemis for denoise/sharpen; Dione for deinterlacing; Gaia for CGI; Rhea for fine texture; Theia for sharpening with noise control; Starlight and Astra for diffusion-based restoration/creative detail).
• Use the Model list to map footage issues to specific codes (e.g., prob-4, iris-3, ahq-12, ddv-3, thd-3, sl-1, ast-2, nyx-3, apo-8, etc.).
• Apply shared parameters (video_type, field_order, auto, focus_fix_level, compression, details, noise, halo, blur, preblur, prenoise, grain, grain_sigma, grain_size, grain_type, recover_original_detail_value) and follow heuristics: start with auto, avoid over-sharpening, reintroduce grain after heavy denoise, and manage diffusion model creativity to limit artifacts.
• Drive the async API lifecycle with: Create POST /video/, Accept PATCH /video/{requestID}/accept, Upload PUT to S3 URLs, Complete-upload PATCH /video/{requestID}/complete-upload, then poll GET /video/{requestID}/status until done. Use per-request X-API-Key, handle 429s with exponential backoff, and segment long sources to respect 500 MB limits. For stills, use the Image API endpoints POST /enhance/async and GET status/download with model and output_format parameters.

When to use it

• For pipelines that generate low-res footage then upscale to 1080p/4K, with the enhancement as a final finishing pass.
• When cleaning AI-generated, compressed, or archival footage; applying denoise, deinterlacing, stabilization, frame interpolation, SDR→HDR, colorization, or foreground removal as needed.
• When a repeatable, automated processing stage is required rather than manual one-off cleanup.

What it can touch

• The Topaz Platform Video API (https endpoints) for async jobs; supported models and codes; video/ and image/ endpoints; per-request limits; available filters and parameters as described above.

Caveats

• Enhancement is a repair/finishing operation and may introduce artifacts if misapplied (over-smoothing, temporal shimmer, face hallucination).
• Do not use for generating new content or for non-Topaz upscalers.
• Model availability and codes can change; verify against live Available Models / OpenAPI schema before hard-coding.
• Pricing is volatile and depends on per-credit costs and plan; diffusion/creative models cost more than deterministic Proteus passes.
• The guidance notes that enhancement should be last in the pipeline and that upscaling after heavy denoise can amplify issues if not addressed first.