Giant Animal Landmark Cuddle · Open Skill 공개 스킬
Giant Animal Landmark Cuddle
Giant Animal Landmark Cuddle은 Digen의 공개 Open Skill입니다. 열어 보면 이 워크플로가 무엇을 만드는지 알 수 있습니다.
Product
0 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/97c3jfhd.jpg
# Giant Animal Landmark Cuddle Turns one animal and one recognizable city landmark into a single vertical photoreal image: the animal is colossal and curls affectionately around the landmark, while the real city around them reads as a tiny physical model. There is no structured project state — everything lives in this conversation and on the canvas. ## When to use - The user asks for a "giant animal hugging / cuddling a landmark", "colossal animal next to a city icon", "miniature city with a huge animal". - The user uploads an animal photo and/or a landmark photo and wants a single merged result image. - The user only describes the animal and the place in words and wants the same style of image. ## Inputs and outputs - Inputs: free text (animal, landmark, optional city name, mood/light) and up to two optional reference images (animal; place or object). - Output: exactly one image asset per request, vertical 3:4, the animal and landmark merged into one coherent scene. ## Steps 1. **Read the request intake rules.** Call `read_skill_file` with `path="references/step-details.md"` and follow its intake checklist and branching table: identify the animal, the landmark, the city name, and which reference images (if any) the user attached. If a reference image is attached, call `recognize_image` on it to read species, markings, landmark architecture, signage and city name instead of guessing. Ask one short clarifying question only if the animal or the landmark is genuinely missing; otherwise proceed with a sensible, clearly-stated choice. 2. **Load the visual rules.** Call `read_skill_file` with `path="references/style-and-prompt.md"`. It holds the mandatory Style / Subject-content / Layout / Must-keep / Must-avoid constraints and the base prompt skeleton. Every prompt you write must satisfy them. 3. **Assemble the prompt.** Call `read_skill_file` with `path="references/prompt-templates.md"` and pick the template that matches the inputs (text only, animal reference only, landmark reference only, both references, object reference). Fill the slots with the concrete animal, landmark, city name and distinctive features gathered in step 1. Write the final prompt in English. Never ship a generic template sentence as the finished prompt — the prompt must name the specific animal, the specific landmark and its readable city name. 4. **Pick the channel and generate.** Use `generate_image`: - No reference image → `model='t2i'`, no `input_urls`. - Any reference image attached → `model='i2i.hd.lite'` and pass the uploaded image URLs in `input_urls` (animal first, landmark/place second). The prompt must name which reference is the animal and which is the place, and must state that the reference identity has to be preserved. - Always pass `orientation='portrait'`, `aspect_ratio='3:4'` (the app's fixed vertical frame), and a short `name` such as `Fox x Big Ben cuddle`. - Pass `entity_type='final_images'`. 5. **Check the result.** Apply the quality checklist in `references/step-details.md` to the returned asset: animal identity preserved, landmark architecture and readable city name intact, miniature surroundings, tilt-shift depth of field, vertical 3:4, no text overlay or watermark. If it fails, follow the retry rules in the same file (one corrective re-generation with the failed point written explicitly into the prompt). 6. **Reply.** Describe the finished scene in one or two natural sentences and mention what the animal, landmark and city name are. Do not paste long URLs. When the image is ready and the user may want a variation, call `set_guidance` in the same turn with options such as a different animal, a different landmark, or a different light/time of day. ## Notes - Failed generation (`success=false`) must be reported honestly; never fabricate an image or an asset link. - Refuse adult or sexualized concepts, gore, or real-person likeness misuse; do not call `generate_image` for those. - Video generation is not part o