upscalerlisted

<objective> Image upscaling utility. Single input image → output upscaled by 2× / 4× / 8× factor. Sharper, higher resolution, with AI-based detail interpolation (vs simple bilinear / bicubic). This skill does NOT: - Generate new content — it ENHANCES existing pixels via super-resolution models - Magic-restore an image from junk-tier (16×16 thumbnail to 4K HD) — input quality matters; garbage in, garbage out - Handle video — one frame at a time only - Vectorize — output remains raster PNG - Remove background (chain with `bg-remover`) - Fix major artifacts (deep blur, motion blur, JPEG mosquito noise at extreme levels) — works best on mildly low-res / soft images </objective> ## ROLE Read input image → call a Replicate-hosted upscaler → save the upscaled output at the requested scale factor. ## PIPELINE 1. **Resolve input**: - `--image <path-or-url>` (required) 2. **Pick provider**: - Default: `nightmareai/real-esrgan` (popular general-purpose, ~$0.005/image at 4×) - Alternatives via `--replicate-model <id>`: - `nightmareai/real-esrgan` — default; best general - `tencentarc/gfpgan` — face-focused (best for portraits) - `jingyunliang/swinir` — alternative general - `philz1337x/clarity-upscaler` — high-fidelity preservation, slower - Other Replicate upscalers: pass via `--replicate-model <slug>` 3. **Scale**: `--scale 2|4|8` (default 4). 4. **Face enhance** (optional): `--face-enhance` enables face restoration (Real-ESRGAN supports GFPGAN-s