# Image Feedback Knowledge

> Use when analyzing, critiquing, improving, editing, or quality-checking images through a closed-loop workflow: define scene-aware judgment standards, set concrete success criteria, modify the image according to those criteria, inspect whether the result meets the criteria, iterate if needed, and output the final image plus rationale. Converts vague aesthetic feedback like "make it premium", "not good-looking", "cheap", "messy", "fake", or "dirty" into industry-benchmarked standards and actionable image-editing plans. Especially useful for people images such as portraits, fashion/editorial images, daily lifestyle images, business headshots, social avatars, and for product, brand, ecommerce, social media, food, interior, event, and portfolio images.

- Skill: `arcdodo/image-feedback-knowledge` (Agent Skill, multi-file: 11 files)
- Install (CLI): `npx skillmds@latest add arcdodo/image-feedback-knowledge`
- Raw SKILL.md: https://api.skillmd.com/api/skills/arcdodo/image-feedback-knowledge/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: arcdodo (https://skillmd.com/u/arcdodo)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/arcdodo/image-feedback-knowledge

---


# Image Feedback Knowledge

Use this skill to turn subjective image feedback into a complete image-improvement loop: standard creation, standard setting, image modification, compliance checking, and final output.

The core rule is scene first:

1. Identify the likely image use case.
2. Recommend 2-4 suitable standards or style directions and ask whether the user has preferred references.
3. Build a judgment standard from the selected or assumed direction.
4. Set concrete success criteria for the current image.
5. Modify the image or produce an editing plan according to those criteria.
6. Check whether the modified image meets the criteria.
7. Iterate if the result misses important criteria.
8. Output the final image or final editing guidance with a concise compliance report.

## Workflow

1. If the user provides an image, inspect the visible content before judging.
2. Before setting standards, recommend 2-4 suitable standards or style directions based on the image and likely use case. Ask whether the user prefers one direction or has their own reference style/brand/creator.
3. If the user explicitly asks to proceed without discussion, choose the strongest direction and state the assumption. If multiple directions would lead to materially different edits, pause for confirmation before editing.
4. Load `references/closed-loop-workflow.md` for the complete standardize-edit-check-output process.
5. Load `references/quick-card.md` for fast mapping from vague words to concrete fixes.
6. Load `references/scenario-benchmarks.md` when the output depends on industry/use-case standards.
7. Load `references/people-image-standards.md` for any people image, especially portraits, headshots, fashion/editorial images, daily lifestyle images, or social avatars.
8. Load `references/diagnosis-checklist.md` when doing a full critique.
9. Load `references/feedback-templates.md` or `references/ai-edit-prompts.md` when the user wants copyable feedback or AI image-editing prompts.
10. If the user asks to build or maintain a case knowledge base, treat that as a separate documentation task rather than part of the image-editing loop. `references/case-training.md` may be consulted as a template, but do not let it expand the current image task.

## Output Shape

For a complete image-improvement task, use this structure:

```text
使用场景：[assumed or stated scene]
推荐风格/标准：[2-4 options, with the recommended one first]
用户参考：[user-provided reference or selected direction]
判断标准：[5-8 criteria with benchmark references]
本次达标要求：[specific pass/fail criteria]
修改计划：[ordered editing actions]
执行结果：[what was changed or what should be changed]
达标检查：[criteria-by-criteria pass/partial/fail]
最终输出：[final image/prompt/advice]
仍需注意：[remaining risks or constraints]
```

For quick critique without actual editing, use:

```text
使用场景：[assumed or stated scene]
对标标准：[top brands/authors/platforms and observable standards]
主要问题：[3-5 concrete image problems]
修改建议：[specific actions]
避免：[over-editing or scene-inappropriate directions]
```

## Judgment Rules

- Do not treat "premium" as a universal style. A premium business portrait, an editorial photo, a lifestyle image, and an Apple-like product visual require different standards.
- Preserve the image's intended function. Do not make daily-life images look like studio ads, or professional headshots look like fashion editorials.
- Prefer concrete language: light distribution, color temperature, saturation, background interference, subject-background separation, skin/material texture, shadow realism, edge artifacts, crop pressure, and visual hierarchy.
- Include constraints whenever relevant: preserve identity, product shape, brand colors, text accuracy, scene authenticity, or lifestyle atmosphere.
- When suggesting benchmarks, describe observable visual standards instead of only naming the brand.
- Before editing, define what success means. After editing, inspect against the same criteria instead of judging by feeling.
- At the first step, actively propose suitable standards/styles instead of asking an empty open-ended question. Example: "I recommend A for brand premium, B for ecommerce clarity, or C for editorial mood. Do you prefer one of these, or do you have a reference brand/creator?"
- If actual image-editing tools are available and the user asks to modify the image, perform the edit, then evaluate the result against the criteria. If image editing is unavailable, output a precise editing prompt and a checklist.

## Reference Index

- `references/quick-card.md`: fastest lookup for vague terms.
- `references/closed-loop-workflow.md`: complete process for making standards, editing, checking, and final output.
- `references/aesthetic-terms.md`: detailed translation dictionary for terms like premium, cheap, dirty, fake, flat, rustic/outdated, no atmosphere.
- `references/diagnosis-checklist.md`: full image diagnosis sequence.
- `references/feedback-templates.md`: reusable modification sentences.
- `references/case-training.md`: case recording and training examples.
- `references/ai-edit-prompts.md`: prompt templates for AI image editing.
- `references/scenario-benchmarks.md`: use-case standards and industry benchmarking.
- `references/people-image-standards.md`: standards for portrait, editorial, lifestyle, business, avatar, and fashion people images.