Red Sun Fashion Editorial · Open Skill 公開技能
Red Sun Fashion Editorial
Red Sun Fashion Editorial 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/dsy7em7d.jpg
# Red Sun Fashion Editorial
Produces one finished **vertical 3:4 fashion campaign poster** in the fixed "Red Sun Fashion Editorial" art direction: crimson sun disc, dry-brush black ink arc, monochrome model, grayscale garment, muted paper-white background, vertical Japanese brush typography plus a serif English masthead.
Text brief and reference images are both optional. There is no structured state — each poster is one image generation, and the poster itself is the deliverable.
## When to use
- The user asks for a "red sun" / Japanese minimalist fashion poster or editorial image.
- The user uploads a model portrait (Image 1) and/or a streetwear garment photo (Image 2) and wants them turned into this poster style.
- The user wants to redo an existing poster with a different garment, headline, model photo or season caption.
## Inputs
| Input | Required | Role |
|---|---|---|
| Text brief | No | Campaign name, headline, season/caption words, gender/pose tweaks |
| Reference image A — model portrait | No | Face, hair, identity of the single model |
| Reference image B — streetwear garment | No | Garment silhouette, fabric, layering, fit |
Both reference images are independent: one, both or neither may be present. Any missing element is invented to match the style, and you say so briefly in the reply.
## Steps
1. **Read the style guide first.** Call `read_skill_file(name='red-sun-fashion-editorial', path='references/red-sun-style-guide.md')` and follow its palette, composition, lighting, typography and prompt templates. Do not improvise the style from memory.
2. **Classify the request** as one of: full poster from text only; poster that personalizes the model photo; poster that personalizes the garment; poster that personalizes both; single-element revision of an existing poster.
3. **Collect the reference images** into `input_urls` in this fixed order: `[model portrait, garment]`. Use the canvas asset URL/id the user just produced, or the uploaded image reference. Pass only the ones that exist — never invent a URL, and never pass the same image twice.
4. **Pick the channel** (write the channel, never a physical model name):
- At least one reference image → `model='i2i'` (image-to-image, reference-driven).
- No reference image → `model='t2i'` (text-to-image).
5. **Write the prompt in English**, in this order, using the templates in the style guide:
1. art direction line (`Japanese-inspired minimalist fashion campaign poster, vertical 3:4 editorial layout`);
2. subject line (single female model, three-quarter or waist-up, hands in pockets, calm confident gaze, face/hair from the portrait reference);
3. garment line (from the garment reference, described as a relaxed monochrome silhouette layered over a white tee);
4. graphic elements (large crimson sun disc centred behind the head with watercolour-bleed edges, one expressive dry-brush black ink arc sweeping across the lower half);
5. background line (off-white paper grain, one narrow grayscale street-photo strip of a lantern-lit rain-wet Tokyo alley on the right edge);
6. typography line (vertical Japanese brush kanji column on the left, two-line serif masthead at bottom left, small letter-spaced caption under it, tiny top-right city/est./year line with hairline rules);
7. light/colour line (flat matte studio light, no hard shadows, palette limited to paper white + ink black + one crimson red);
8. `Avoid:` clause (bright saturated colours, colour garment, busy background, 3D-render look, extra people, distorted limbs, garbled or invented text).
Keep the poster's own text strings exactly as specified — quote them in the prompt so they are legible and not gibberish.
6. **Call the image tool** exactly once per poster:
```
generate_image(
prompt=<English prompt from step 5>,
input_urls=[<references from step 3, if any>],
model='i2i' # or 't2i' when no reference image is given
orientation='portrait',
aspect_ratio