seahi/baoyu-skills
1
0
Fork 0
mirror of https://github.com/JimLiu/baoyu-skills.git synced 2026-08-07 17:33:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix(baoyu-cover-image): enhance reference image analysis and prompt generation

Browse Source 
- Add deep analysis template for extracting specific visual elements
- Require MUST INCORPORATE section with concrete, reproducible details
- Add integration approach for spatial layout instructions
- Clarify ref-capable backend requirements (Google/OpenAI)
This commit is contained in:
Jim Liu 宝玉
2026-02-06 16:07:09 -06:00
parent 86a84739e8
commit 6fffe252c0
2 changed files with 87 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files
skills/baoyu-cover-image/SKILL.md
+33 -12
View File
@@ -317,16 +317,23 @@ Reference Style Extracted (no file):
- Identify visual metaphors
- Detect content type
**1.3 Reference Image Analysis** (if provided in Step 1.0)
**1.3 Reference Image Deep Analysis** ⚠️ CRITICAL (if provided in Step 1.0)
For each reference image:
References are high-priority inputs. Perform deep visual analysis to extract **specific, concrete, reproducible** elements — not vague summaries.
| Analysis | Description |
|----------|-------------|
| Visual characteristics | Style, colors, composition |
| Content/subject | What the reference depicts |
| Style match | Which type/palette/rendering align |
| Usage recommendation | `direct` / `style` / `palette` |
For each reference image, extract ALL of the following:
| Analysis | Description | Example (good vs bad) |
|----------|-------------|----------------------|
| **Brand elements** | Logos, wordmarks, specific typography treatments | Good: "Logo uses vertical parallel lines for 'm'" / Bad: "Has a logo" |
| **Signature patterns** | Unique decorative motifs, textures, geometric patterns | Good: "Woven intersecting curves forming diamond grid" / Bad: "Has patterns" |
| **Color palette** | Exact hex values or close approximations for key colors | Good: "#2D4A3E dark teal, #F5F0E0 cream" / Bad: "Dark and light colors" |
| **Layout structure** | Specific spatial arrangement, banner placement, zone splits | Good: "Bottom 30% dark banner with branding" / Bad: "Has a banner" |
| **Typography** | Font style, weight, spacing, case, positioning | Good: "Uppercase, wide letter-spacing, handwritten ep number" / Bad: "Has text" |
| **Content/subject** | What the reference depicts | Factual description |
| **Usage recommendation** | `direct` / `style` / `palette` | Based on analysis |
**Output format**: List each element as a bullet point that can be directly copy-pasted into a prompt as a mandatory instruction.
**1.4 Language Detection**
- Detect source language
@@ -380,10 +387,15 @@ references:
| Situation | Frontmatter | Body |
|-----------|-------------|------|
| Reference file saved to `refs/` | Add to `references` ✓ | Brief style note |
| Reference file saved to `refs/` | Add to `references` ✓ | **Detailed mandatory style instructions** (see below) |
| Style extracted verbally (no file) | Omit `references` | Full style description |
| File in frontmatter but doesn't exist | ERROR - fix or remove | — |
⚠️ **"Brief style note" is NOT sufficient.** When references are provided, the prompt body MUST contain a dedicated `# Reference Style — MUST INCORPORATE` section with:
1. Per-reference breakdown of **specific visual elements** extracted in Step 1.3
2. Each element prefixed with "MUST" or "REQUIRED"
3. An **integration approach** describing exactly how reference elements should be composed into the cover layout (e.g., "Use a split layout: illustration area 65% + dark branded banner 35%")
### Step 4: Generate Image
**4.1 Backup existing** `cover.png` → `cover-backup-YYYYMMDD-HHMMSS.png` (if regenerating)
@@ -421,8 +433,8 @@ references:
| Skill Supports `--ref` | Action |
|------------------------|--------|
| Yes (e.g., baoyu-image-gen with Google) | Pass reference images via `--ref` |
| No | Convert to text description, append to prompt |
| Yes (e.g., baoyu-image-gen with Google/OpenAI) | Pass reference images via `--ref` |
| No | If usage is `direct`, switch to a ref-capable backend (baoyu-image-gen + Google multimodal or OpenAI GPT Image edits). If unavailable, convert to text description and remove direct refs |
**Verification**: Before generating, confirm reference processing:
```
@@ -434,7 +446,7 @@ Reference Processing:
**4.4 Generate**
1. Call selected skill with prompt file path, output path (`cover.png`), aspect ratio
2. If references with `direct` usage AND skill supports `--ref`: include `--ref` parameter
2. If references with `direct` usage: use ref-capable backend and include `--ref` (recommend `baoyu-image-gen --provider google --model gemini-3-pro-image-preview` or `--provider openai --model gpt-image-1.5`)
3. On failure: auto-retry once before reporting error
### Step 5: Completion Report
@@ -483,6 +495,15 @@ All modifications automatically backup existing `cover.png` before regenerating.
- `--no-title` is alias for `--text none`
- `--style` presets are backward-compatible; explicit `--palette`/`--rendering` override preset values
### Reference Image Priority ⚠️
When user provides reference images, they are **HIGH PRIORITY** and MUST strongly influence the generated cover:
- **References override defaults**: If reference images conflict with preferred palette/rendering, the reference visual identity takes precedence in the prompt
- **Concrete > abstract**: Extract specific, reproducible visual elements (exact patterns, logo treatments, color hex values, layout structures) — not vague descriptions like "clean style"
- **Mandatory language**: Use "MUST", "REQUIRED", "CRITICAL" in the prompt for reference elements — generic "follow the style" is insufficient for image generation models
- **Visible in output**: After generation, verify that reference elements are actually present in the cover. If not, strengthen the prompt and regenerate
### Composition Principles
- **Generous whitespace**: 40-60% breathing room; avoid cluttered layouts