Misty Mountain Adventure Portrait · Open Skill 公開技能
Misty Mountain Adventure Portrait
Misty Mountain Adventure Portrait 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Other
1 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/cmuq4qgp.jpg
# Misty Mountain Adventure Portrait Produces one cinematic still image: a hyper-realistic bust-to-three-quarter portrait of a subject on a misty mountain ridge, in fog, with cool daylight and sharp natural skin detail. Its signature behaviour is the **outfit swap**: when an identity photo is uploaded, the face (and other identity anchors) is kept while wardrobe, pose, props and mountain scene change; the light, color grade and photographic treatment stay consistent so a series looks like one shoot. ## When to use - The user asks for a misty mountain portrait, alpine/adventure portrait, "mountain fog portrait", or a "cinematic still of me in the mountains". - The user uploads a photo of a person and asks for a costume / outfit / scene change while keeping the same face. - The user asks to tweak light, fog density, color, framing, or camera feel of a portrait produced earlier in the conversation. ## Inputs - **Text**: subject description, optional wardrobe / theme / pose / weather hints. - **Optional reference image**: the identity photo (an asset in this conversation). Identity photo is optional — without it the skill still produces a text-to-image portrait. ## Steps 1. **Parse the request into four slots**: `SUBJECT` (who), `WARDROBE` (theme + garments), `POSE`, `ENVIRONMENT` (fog density / terrain / time of day). Defaults when a slot is missing — subject: an adult man with rugged outdoor outerwear; wardrobe: the *Alpine Trekker* set; pose: three-quarter turn, gaze off-camera to the right; environment: pine-ridge overlook under thick valley fog, overcast daylight. 2. **Load the look before the first render of a conversation**: read `references/style-guide.md`. It holds the fixed global blocks (framing, directional light, color grade, camera/render treatment), the fill-in prompt template, and the negative list. When the request involves an outfit/pose/scene swap, or the user is building a series, also read `references/outfit-themes.md`. 3. **Handle the identity photo (if present)**: pass its asset id in `input_urls` and copy the identity-anchor sentence from `references/outfit-themes.md` into the prompt. Only if you need the anchors spelled out in words (or the asset cannot be fed to the generator) call `recognize_image(image_url=<identity asset_id>, prompt="Describe this person's face, hair, glasses, facial hair, skin tone and apparent age as identity anchors for a portrait.")` and use its answer to write the anchors, never to invent new features. 4. **Compose the English prompt** using the template in `references/style-guide.md`: fill every block (style, light, subject/identity, wardrobe, pose, environment, grade, camera, render) and append the negative list as a final sentence beginning `Avoid: ...`. Write one single paragraph — `generate_image` takes one `prompt` string, there is no separate negative field. 5. **Call `generate_image`**: - with identity photo: `generate_image(prompt=..., input_urls=[<identity asset id>], model='i2i', orientation='portrait', name='<Theme> Portrait', entity_type='characters')` - text only: `generate_image(prompt=..., model='t2i', orientation='portrait', name='<Theme> Portrait', entity_type='characters')` - add `aspect_ratio='3:4'` only when the user explicitly asks for the reference card's exact 3:4 still ratio. 6. **Check the result**: if the person's face drifted, the frame became a wide landscape shot, or the fog/grade collapsed into a warm golden-hour look, re-run once with the same identity anchor and a stronger identity + light sentence. Never silently pass off a failed identity match as success. 7. **Reply** in one or two natural sentences (subject, theme, look) and offer 2–3 concrete next steps in the same turn with `set_guidance`, e.g. "Switch to the Winter Expedition outfit", "Push the fog thicker and light colder", "Same look, lower the crop for the whole jacket". ## Parameter conventions | Parameter | Value | |---|---| | `model` | `'i2i'` when an identity photo i