---
name: upscale
description: "Router for every upscale job — image or video. Classifies the input (close-up, full figure, character sheet, film still, start frame for video, regeneration, skin, finished video, video takes) and the priority (speed / detail / fidelity), then hands off to the matching upscale-* sub-skill with exact settings and time. Use when the user says upscale, 4K, sharpen, enhance, restore detail, Magnific, Starlight, higher resolution, апскейл, увеличить, 4К."
---

These creative references describe source experiments with external tools. Use your own supported account and tools; this download installs no automation. Settings and timings are experimental observations, not current FlashBoards availability.

# upscale · meta-skill (router)

**Work for the person's own character: their photo / look sheet / voice from their character card. The test images cited below (creation ids) belong to a demo character (Mara) — never reuse them for a person.**

Every recommendation here is measured on source experiments (29.09.2026: a demo character sheet (Mara), film stills, MiniMax H3 videos).
Article with all the evidence: https://flashboards.yaroflasher.com/learn/post-production/ai-upscaling/

## 1 · Classify the job

| Input | Go to |
| --- | --- |
| A face close-up / portrait | `upscale-closeup` |
| A full figure, a wide shot, small faces | `upscale-fullbody` |
| A character sheet (close-up + full body in one image) | `upscale-sheet` |
| A film still, screenshot, small reference | `upscale-film-still` |
| A frame that will start a video | `upscale-start-frame` |
| "Just make it bigger, fast" — pixel-exact fidelity not needed | `upscale-regenerate` |
| Only the skin of a big face | `skin-enhance` |
| A finished, edited video | `upscale-video-final` |
| Video takes in progress | `upscale-video-takes` (do NOT upscale takes) |

## 2 · Choose by quality and fidelity

Choose the input type and required fidelity first. Use regeneration only when a redraw is acceptable; use the tested Magnific presets when identity and contours must stay close to the source. The sub-guides retain measured timings.

## 3 · How the skills link — the full pipeline

```
generate the frame at the right size ─── upscale-start-frame
        │ (already have a 1K image?)
        ▼
upscale the still ── upscale-closeup / upscale-fullbody / upscale-sheet / upscale-film-still
        │ (big face only, optional)
        ▼
skin ────────────── skin-enhance
        ▼
video takes ─────── upscale-video-takes   (generate at final resolution, keyframes far off the timeline)
        ▼
edit, then ONE final pass ── upscale-video-final
```

## 4 · Global rules (break none)

0. **Pick the scale by the input size** (source checks, 29.09.2026):
   - input ~1200 px and more (a character sheet, a big frame) → **Magnific Ultra Sublime · 2× · Grainy Analog** — the best; 4× of the same image **downgrades the look** (4× Grainy, Balanced, Portraits all checked).
   - a small standalone 1K image, ~700 px (one close-up, one full figure) → **Ultra Sublime · 4× · Portraits** (15–34 s) — the sharpest face we got (s7z7eN7l8e, TdFBgcMVNR on the demo sheet).
   - **a character sheet → upscale the close-up and the full figure as separate images**, 4× Portraits each (`upscale-sheet`): both halves come out clearly sharper than the sheet in one pass.
0. **Same settings, another run = the same face, different fine texture.** For the best still run 2–3 times and pick.
0. **Blurring the input first does not help Ultra Sublime** — it keeps the blur (1.5 px: a little softer; 3 px: stays blurry).
0. **Focus on skin → `skin-enhance`** (Skin Enhancer Creative) — big faces only.

1. **Fix the character before upscaling.** Magnific is faithful to mistakes too (1K brown eyes stayed brown).
2. **Skin Enhancer only on big faces.** On a full figure every version ruined the face (pitted skin, wrong eye colour).
3. **GPT Image 2.5 at 4K paints on close-ups** (brush-stroke brows, smeared fabric). Close-ups at 2K, 4K only for wide shots.
4. **quality high added no visible detail over medium** in the measured generation comparisons. Exception: GPT used as an upscaler — high halves the drift.
5. **A plain resize adds pixels, not detail** (video from a resized frame: 1.04× — noise).
6. **Magnific Creative Vivid / Wild change the person** — never on a character.
7. **Upscale the final cut, not the takes**.

## 5 · Settings and times (measured)

| Tool · setting | Time |
| --- | --- |
| **Magnific Upscaler · Ultra Sublime 2× · Grainy Analog — the best on sheets and big frames** | ~40–70 s |
| **Magnific Upscaler · Ultra Sublime 4× · Portraits — the best on a small 1K close-up or full figure** | 15–34 s |
| Magnific Upscaler · Ultra Sublime 2× · Balanced — cleaner, less texture | 41 s |
| Magnific Upscaler · Ultra Sublime 4× on a sheet / 1200 px frame — looks worse than 2× | 28–46 s |
| Magnific Upscaler · Ultra Sublime 4× · Balanced on film stills — softer, closer to the source than Portraits | under 35 s |
| Magnific Upscaler · Creative 2× · Subtle | ~48 s |
| Magnific Skin Enhancer · Creative / Faithful / Flexible | 62–133 s |
| Magnific video · Topaz Starlight Precise 2.6 → 3840 px | 16 min |
| GPT Image 2.5 Flare · low · 2K / 4K | 19 s / 35 s |
| Nano Banana 2 · 4K | 43 s |
| MiniMax H3 Max Turbo i2v · 768P / 1080P | ~9 s per 15 s + queue |