Halloween in the Basement · Open Skill 공개 스킬
Halloween in the Basement
Halloween in the Basement
Auto
1 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/jnkushqa.jpg
# Halloween in the Basement Turn a short text description from the user into a single image on the "Halloween · basement" theme. There is no structured state: each image is one `generate_image` call, and the result goes to the canvas automatically. ## When to Use - The user wants an image of "Halloween in a basement" / a Halloween cellar / a Halloween atmosphere scene. - The user gives a piece of text (in any language) and wants it drawn as a Halloween basement picture. - The user asks to modify the previous result (change the shot, add props, change the lighting, switch to landscape or portrait, "one more version") → just call the tool again with the new requirements; do not re-ask about the background. ## Steps ### 1. Rewrite the user's description into a single English prompt User input is often just a few words (for example, "Halloween in the basement"). In that case, fill in the gaps using the **default scene** below: two characters in Halloween costumes, carving pumpkins by torch / candle light in an old basement. Build the prompt in this order (make each item concrete — use specific nouns, not just abstract adjectives): 1. **Subject and action** — whatever subject the user described; if the user gave none, use `Two figures in Halloween costumes carving pumpkins`. 2. **Environment** — the basement: `inside an old stone basement / cellar`, optionally adding `bare brick walls, low ceiling, damp floor`. 3. **Halloween decorations and props** — `carved jack-o'-lanterns with glowing faces, scattered pumpkin seeds and pulp, dripping candle stubs, cobwebs in the corners, dusty wooden crates, old shelves, hanging bare bulbs, bats, dried leaves`. Replace or extend these with any props the user names explicitly. 4. **Light source (always include)** — `a wall torch / candle flames casting a flickering warm orange key light, deep cool blue-green shadows`. 5. **Color palette** — `deep purple, burnt orange, sickly green, charcoal black`. 6. **Mood** — `spooky, eerie, cozy-creepy, festive Halloween atmosphere`. 7. **Composition and camera** — pick one of medium shot / wide shot / low angle, e.g. `medium shot, slight low angle, cinematic lighting, shallow depth of field`. 8. **Quality terms** — `highly detailed digital illustration, volumetric light`. A complete example (usable directly when the user only says "Halloween in the basement"): ```text Two figures in Halloween costumes carving pumpkins inside an old stone basement, surrounded by carved jack-o'-lanterns with glowing faces, scattered pumpkin seeds, dripping candle stubs and a wall torch casting a flickering warm orange light, dusty wooden crates, cobwebs, hanging bare bulbs, bats, deep purple and burnt orange color palette, deep shadows, spooky festive Halloween atmosphere, medium shot, slight low angle, cinematic lighting, highly detailed digital illustration ``` ### 2. Call `generate_image` ```text generate_image( prompt="<the English prompt from step 1>", model='t2i', orientation='landscape', # landscape 16:9 by default name='Halloween in the Basement', entity_type='halloween_scenes' ) ``` - **`model` is always the channel `t2i`** (text-to-image). Switch to `model='i2i'` with the reference image in `input_urls` only when the user uploaded a reference image and asks to "modify based on this one". Never write a physical model name. - `orientation`: `landscape` by default; pass `portrait` when the user wants a poster / phone wallpaper / vertical image; pass `square` for an avatar / square image. - `resolution`: omit it (defaults to 720P); pass `1080P` only when the user explicitly asks for high definition. - Do not pass `aspect_ratio`, `batch_size`, or `width` / `height`. - Use a short English title for `name` (e.g. `Halloween in the Basement`, `Pumpkin Carving Night`) so it displays well on the canvas. - `entity_type` is always `halloween_scenes`. ### 3. When multiple options are needed Each call produces only 1 image. If the user wants "a few