Halloween Night Parade Costume Video · Open Skill 公開スキル
Halloween Night Parade Costume Video
Halloween Night Parade Costume Video
Custom
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-15/euvfkr2f.jpg
# Halloween Night Parade Costume Video
Turn a **single uploaded reference image** into a vertical Halloween short video: the subject wears Halloween role-play accessories, walks down a street where jack-o'-lanterns light up one by one, and finally turns around and smiles.
The input is the reference image; the output is a **single-shot** short video (not a multi-stage project, no structured state).
## When to Use
- The user gives an image and wants "Halloween costume / Halloween mood video / jack-o'-lantern street / Halloween main character arrival".
- The user is unhappy with the last version and wants different accessories, a different street mood, or a different camera feel, then another generation.
- The user only says "make me a Halloween video" but has not provided an image yet: ask for an image first; never invent a subject out of thin air.
## Steps
1. **Confirm the reference image.** Take the `asset_id` or URL of the user's uploaded image; when there are several, pick the one where the subject is clearest and occupies the largest area. If there is no image, ask the user for one.
2. **Understand the subject.** Call `recognize_image(image_url=<reference image>)` to get the subject type (person / pet / still-life product / illustrated character) and the features that must be preserved: hairstyle and hair color, facial or shape features, main clothing color, pose, and approximate framing ratio. These features must go into the prompt.
3. **Assemble the English prompt** (template and conventions in the next section), substituting the concrete features from step 2 into the placeholders.
4. **Generate one shot with a single call**:
```
generate_video(
prompt=<the English prompt from the previous step>,
input_urls=[reference image],
model='ref2v',
orientation='portrait',
name='Halloween Night Parade',
entity_type='final_videos'
)
```
- `model` is always the channel `ref2v` (reference-to-video, preserves subject consistency); do **not** pass a physical model name, and do **not** pass `gen_mode` — the modality is decided by the channel.
- `input_urls` needs at least 1 reference image (the 1st one is the subject reference).
- Default to vertical `orientation='portrait'`; switch to `landscape` only when the user explicitly asks for a landscape video.
- Do not proactively pass `duration` / `resolution`; use the platform defaults (out-of-range values are clamped automatically).
- One call = one shot; wait for this call to return, and do not poll or resubmit the same task.
5. **Reply.** On success, summarize the scene and mood in one or two natural-language sentences (subject + accessories + jack-o'-lanterns lighting up one by one + the closing turn-and-smile), and do **not** paste long URLs; the output goes to the canvas automatically. In the same turn you may call `set_guidance` with 2–3 next-step suggestions (for example "redo it with a misty graveyard background", "give the subject a pumpkin hand lantern"). On failure, explain the reason honestly and do not invent a video link.
## Prompt Conventions
**Language**: the prompt is written in English; keep any dialogue supplied by the user in its original language.
**Template** (replace `<>` with the concrete features recognized in step 2):
```
A cinematic vertical 9:16 shot: the subject from the reference image — <hairstyle/hair color/facial and shape features/main clothing color, kept unchanged> — dressed up for Halloween with small costume accessories (<person: pointed witch hat + short cape + pumpkin brooch; pet: tiny wizard cape + mini pumpkin hat; still-life/product: only add Halloween staging such as jack-o'-lanterns and cobwebs, without changing the object itself>), walking down a festive Halloween street at night. Street props: carved jack-o'-lanterns lining the road, glowing paper lanterns and string lights floating overhead, cobblestone pavement scattered with dry autumn leaves, cobwebs and bat silhouettes a