image-outpainting-output · Open Skill 公開技能

image-outpainting-output

image-outpainting-output 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。

@pendenskull

Auto

0 0 1

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/571213/2026-10-05/j9bkpnkn.jpg

# Image Outpainting Output Skill

## Purpose

Use this skill when the user wants to expand an existing image into a new aspect ratio, create multiple campaign formats from one source, or prepare outpainted assets for ads, social media, banners, stories, thumbnails, product pages, or other layouts.

The core objective is **not merely to generate a larger canvas**.

The objective is to produce an output that is:

1. Technically valid.
2. Visually coherent.
3. Faithful to the important subject/product.
4. Useful for the intended layout.
5. Ready for downstream design or export.

Treat every outpaint as a new visual composition that requires review.

---

# 1. OUTPUT-FIRST PRINCIPLE

Always optimize for the final deliverable.

Do not report:

> "Outpainting completed."

Report whether the resulting asset is actually usable.

The output decision must answer four separate questions:

### A. Technical validity
- Did generation succeed?
- Is the image readable?
- Is the requested aspect ratio correct?
- Are the decoded dimensions correct or acceptable?
- Is the requested output format correct?

### B. Subject / product fidelity
- Is the main subject still the same?
- Are important shapes preserved?
- Are brand marks, labels, packaging, logos, facial identity, or defining details intact?
- Did the model add duplicate products or objects?
- Did the model unintentionally redraw the original subject?

### C. Scene continuity
- Does the generated area continue the original scene naturally?
- Are perspective lines continuous?
- Is lighting direction consistent?
- Are shadows, reflections, textures, patterns, and gradients plausible?
- Are seams visible at delivery size?

### D. Layout usefulness
- Is there usable negative space?
- Is the subject positioned well for the destination?
- Is there room for headline, CTA, logo, UI overlays, or campaign graphics?
- Does the output still work after the planned text/design layer is placed?

A technically correct image can still fail the task.

---

# 2. INPUT CONTRACT

Before processing, identify:

- Source image.
- Source dimensions.
- Source orientation.
- Intended destination aspect ratio(s).
- Intended use: ad, story, banner, poster, thumbnail, catalog, etc.
- Main protected subject.
- Important details that must not change.
- Preferred output format.
- Planned text / logo / CTA area.
- Any destination safe-area requirements.

Keep an untouched source master.

Normalize orientation before measuring width and height.

Record the working source dimensions.

When practical, record a source revision identifier or checksum so later outputs can be traced to the exact input.

---

# 3. DECIDE: CROP, CANVAS, OR OUTPAINT

Do not use outpainting automatically.

Choose the smallest transformation that achieves the destination.

### Use a crop when:
- The source already contains enough context.
- Important content remains intact.
- No generated scenery is needed.

### Use a designed canvas when:
- A flat background, gradient, color block, or graphic panel can solve the composition.
- Precise text placement matters more than photographic continuation.

### Use outpainting when:
- The destination requires more surrounding scene.
- Natural continuation of the environment is important.
- Cropping would remove commercially or narratively important content.
- Stretching would distort the subject.

Do not use AI-generated pixels where ordinary layout can solve the problem more reliably.

---

# 4. DESTINATION BRIEF

For every output, create a small destination brief before generation.

Use:

```text
OUTPUT NAME:
INTENDED USE:
TARGET ASPECT RATIO:
TARGET / EXPECTED DIMENSIONS:
MAIN SUBJECT:
PROTECTED DETAILS:
DESIRED SUBJECT SCALE:
DESIRED SUBJECT POSITION:
TEXT / LOGO ZONE:
UI / SAFE AREA:
BACKGROUND CONTINUATION:
ACCEPTABLE VARIATIONS:
REJECTION RULES:
OUTPUT FORMAT:
```

The brief must include explicit rejection rules.

Example:

```text
Reject if:
- product logo changes
- product count changes
- mug handle disa