Image Upscaler & Enhancer · Open Skill 公開スキル

Image Upscaler & Enhancer

Image Upscaler & Enhancer は Digen 上の公開 Open Skill です。開くと、このワークフローが何を作るか分かります。

@waytitiai

Custom

14 1 11

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4210253/2026-10-18/qqbfgtyy.png

---

allowed-tools:
  - generate_image
  - recognize_image
---

# Image Upscaler & Enhancer

## When to use
- The user wants to upscale / enhance an existing image (photo, illustration, sketch) — sharper, more detailed skin texture, body, and hair, up to 2K resolution.
- The result must keep the original subject's facial identity and pose.
- Supports NSFW/adult images (bare `i2i` channel only — never the `.hd` variants, which block adult content).
- The user wants any of the additional features below: batch upscale, face/portrait beautify, color correction, denoise/deblur, remove watermark/text/object, outpaint/extend, crop to a ratio, area-specific detail enhancement, style transfer, or a specific target resolution.
- If the source is a rough sketch or a crude manipulation (not a clean photo), offer the user the option to repair/fix that manipulation or sketch into a photorealistic image (see the dedicated section below).
- If the source is a photoshopped/composited image with non-seamless joins (e.g. a face pasted onto a body with visible crop edges, mismatched skin tone, or hard seams), repair those joins so the result looks like one natural photo (see the seam-repair section below).
- If the source is an old, blurry, grainy, or low-quality photo (vintage/old photograph), restore it into a clear, high-resolution modern photo (see the vintage-restoration section below).

## Base upscale flow (used by most features)
1. Obtain the source image: if the user uploaded one, use its `asset_id`; if they reference a library item, use `search_library` / `get_library_item` to get its media `uri`.
2. Call `recognize_image(image_url=<source>)` to capture the subject's facial features, pose, body, hair, and any visible seams, crop edges, or mismatched skin tone from compositing. This analysis drives identity preservation.
3. Call `generate_image` with `model='i2i'` (bare `i2i` — supports NSFW; do **not** use `i2i.hd` / `t2i.hd` / `.hd.lite`, which block adult content), `input_urls=[source image]`, and a prompt describing: upscale to higher resolution, enhance skin texture, detailed body and hair, and keep the subject's exact facial identity and pose. **Preserve the source image's original aspect ratio — do not force a fixed orientation.** Derive the source's aspect ratio from its dimensions and pass the matching `aspect_ratio` (e.g. `'3:4'`, `'4:5'`) when it is non-standard; otherwise pass the `orientation` that matches the source. Set `resolution` to `2K` by default (see the target-resolution feature below for choices) — the platform auto-clamps down if the model cannot reach it.
4. Pass `name="Enhanced <subject>"` and `entity_type="enhanced_images"`.
5. Reply with a natural-language summary of what was enhanced; the enhanced image appears on the canvas.

## Feature: Natural skin texture (skin pores) — apply to every enhancement
- Always instruct the model to preserve and enhance realistic skin texture with visible natural pores and fine skin detail, and explicitly avoid an over-smoothed, waxy, plastic-like "doll" look.
- Add this line to every enhancement/beautify/restoration prompt: "Preserve and enhance realistic skin texture with visible natural pores and fine skin detail; do NOT over-smooth the skin — avoid a waxy, plastic, or airbrushed appearance."
- This is a default behavior of this skill, not an optional extra — apply it in every `generate_image` prompt unless the user explicitly asks for a heavily smoothed look.

## Feature: Batch upscale
- When the user provides multiple images, process each one through the base upscale flow with a separate `generate_image` call per image, keeping each source's own aspect ratio.
- Pass `name="Enhanced <subject> N"` (N = the image's position in the batch) and `entity_type="enhanced_images"` for each.
- Summarize the batch in the final reply (how many images, what was enhanced).

## Feature: Face / portrait beautify
- When the user wants to beautify a face/portrait, add to the prompt: "Enhance