| name | cover-image |
| description | Create Chinese video cover images and thumbnails from a reference image, especially for AI/tech tutorial covers requiring exact Chinese title text, soft-light 3D-rendered portrait aesthetics, and exact 16:9, 4:3, and 3:4 aspect ratios. Use when the user asks to make, redo, or adjust cover images/thumbnails with a supplied reference image and Chinese text. Treat 3:4 as a strict ratio that must never be substituted with 2:3. |
Cover Image
Core Intent
Create polished Chinese video cover images from a supplied reference image while preserving the reference's soft-light, high-precision 3D-rendered portrait feel. Prioritize accurate text, usable thumbnail composition, and direct delivery in the Codex conversation.
Required Skill Chaining
When generating or editing raster cover imagery, use $imagegen first and follow its built-in tool workflow. Treat the user-provided image as the primary reference or edit target.
If the image model produces inaccurate Chinese text, use deterministic local post-processing to cover and redraw the exact requested text with an installed Chinese-capable font. Do not accept misspelled, duplicated, or substituted title text as final.
Delivery Rules
- Never write generated images into this skill directory, including
cover-image/, or any subdirectory of it.
- Never use a path relative to this skill directory, such as
cover-image/cover-image.png, for generated outputs.
- Never choose an output folder or filename on the user's behalf.
- Keep generated images in the Codex conversation by default.
- Only save image files when the user explicitly provides a concrete destination path outside this skill directory.
- A request to "generate", "make", "show", "deliver", or create "downloadable" cover images is not permission to save files.
- If the user asks to save but does not provide a concrete destination path, ask for the path before saving.
- If saving is requested with an explicit destination, write only to that destination and report the saved paths.
- If no explicit save destination is provided, do not create files and do not report a file path.
Default Cover Requirements
Use these defaults unless the user says otherwise:
- Text: two title lines, exactly as provided by the user.
- For the recurring AI tutorial cover pattern, use:
- Aspect ratios: create exactly three covers by default: one 16:9 cover, one 4:3 cover, and one 3:4 cover.
- Hard ratio constraint:
3:4 means width:height = 3:4. Never generate, crop, resize, label, or deliver a 2:3 image as the 3:4 cover. These ratios are not interchangeable or acceptable approximations.
- Exception: create only one cover when the user explicitly asks for a single image or specifies exactly one aspect ratio.
- Suggested output sizes:
- 16:9:
1920x1080
- 4:3:
1600x1200
- 3:4:
1200x1600
- Output format: PNG only when the user explicitly requests saved files with a concrete destination path; otherwise deliver images directly in the conversation without creating files.
Visual Direction
Preserve the reference image's aesthetic before adding cover polish:
- Keep soft daylight, gentle highlights, and natural skin tones.
- Avoid harsh contrast, action-poster lighting, smoky dark grading, heavy black shadows, and over-saturated orange/blue color casts.
- Keep the image bright enough to feel close to the reference, not dramatically darker.
- Preserve the subject's face identity, hair detail, clothing color, and premium realistic 3D-rendered texture.
- Use subtle readability treatment behind text: soft shadow, light stroke, gentle darkening, or local blur only where needed.
Composition Rules
Protect the face. Text may overlap the body, jacket, shoulders, chest, lower hair, or background, but must not cover the eyes, nose, mouth, or main face area.
For 16:9:
- Put the face on the right or center-right.
- Use left or lower-left space for the two-line title.
- Keep title spacing generous and readable at thumbnail size.
For 4:3:
- Avoid cramped title blocks.
- Let the title occupy lower-left or lower-middle space and overlap body/clothing if needed.
- Prefer comfortable line spacing and margins over forcing all text into a narrow side column.
For 3:4:
- Lock the canvas to an exact
3:4 width-to-height ratio, preferably 1200x1600. Do not use 2:3 dimensions such as 1024x1536 or 1200x1800.
- Keep the face in the upper half.
- Put the title in the lower third or lower quarter.
- Allow text to cover the jacket/body but not the face.
Prompt Pattern
Use a structured prompt like this, adapting only the user-specific text and ratios:
Use case: ads-marketing
Asset type: Chinese video cover thumbnail, <aspect ratio>.
Canvas ratio: use the exact requested width:height ratio. For 3:4, use an exact 3:4 canvas such as 1200x1600; never use or substitute 2:3.
Input image: use the provided image as the primary visual reference. Preserve the same realistic ultra-detailed 3D-rendered subject, face identity, soft skin, detailed hair, clothing, and gentle soft daylight from the reference image.
Primary request: Create a polished Chinese video cover with a bright, soft look. Avoid harsh contrast, smoky dark grading, heavy black shadows, and action-poster lighting. Keep the palette close to the reference: soft neutral daylight, warm skin tones, airy background, refined high-precision 3D render feeling.
Text (verbatim, two lines only):
"<line 1>"
"<line 2>"
Typography: bold modern title type, large and readable; first line white, second line warm gold or another user-approved accent; subtle outline/shadow only enough for readability.
Composition/framing: <ratio-specific placement>. Text may overlap body/clothing, but must not cover eyes, nose, mouth, or the main face area.
Constraints: Chinese text must be exactly the two lines above, no misspellings, no extra text, no watermark. Aspect ratio must be exact; 3:4 must never become 2:3.
Validation
Before finishing:
- Inspect each image visually.
- Confirm the text is exact and contains no extra text.
- Confirm lighting remains soft and not significantly darker than the reference.
- Confirm text does not cover the face.
- Confirm that the default output set is three images: 16:9, 4:3, and 3:4, unless the user explicitly requested one image.
- Verify the actual pixel dimensions of every output, including images delivered only in the conversation. For the 3:4 cover, confirm
4 × pixel width = 3 × pixel height (for example, 1200x1600). Reject and regenerate any 2:3 result; do not relabel or crop it loosely as 3:4.
- Confirm that no files were created unless the user provided a concrete save destination outside this skill directory.
- If files were saved because the user provided an explicit destination path, confirm dimensions with an image tool such as
sips -g pixelWidth -g pixelHeight and report the saved paths and dimensions.
- If files were not saved, present the generated images directly in the conversation and identify their aspect ratios.