Halloween Video Generator · Open Skill 公開技能
Halloween Video Generator
Halloween Video Generator 是 Digen 上的公開 Open Skill。開啟後即可了解此工作流能製作什麼。
Beauty
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/zsdk3gc3.jpg
# Halloween Video Generator Turn one line of user text (optionally with 1–3 reference images) into a cinematic 3D Halloween video. One call produces one shot; call again if the user wants more variants. There is no structured project state — write the creative setup in your reply and reference already generated assets by their canvas URL. ## When to use - The user wants a Halloween-themed video / short film / Halloween promo - The user provided reference images (lead character / second character / story location) and wants them kept consistent - The user only gave one line of idea (jack-o'-lantern, candy thief, Halloween date, etc.) and you need to fill in the style and imagery - The user only wants a single Halloween-themed still frame / cover image ## Input conventions - **Reference images (optional, at most 3)**: read them in order as ① lead character ② second character ③ story location. Fewer than 3 is fine; if the user did not state the order, assume this order and explain your assumption in the reply. - **Text**: one line of story or visual idea. If the user gave none, use the default fantasy scene: **a jack-o'-lantern by the roadside suddenly lights up and the character looks straight at it**. ## Steps ### 1. Write the prompt (English) Rewrite the user's idea into one English prompt and **spell out each of the following four conventions**: - **Style / atmosphere (Halloween)**: decorations, props and lighting all stay on the Halloween theme — jack-o'-lanterns, candlelight, drifting mist, bare branches, spider webs, scattered candy; warm orange pumpkin firelight against deep purple and blue night; volumetric light, rim light. - **Ghost theme (restrained)**: add only a **small** amount of ghost detail — one or two translucent, faintly glowing spirit silhouettes in the background; they must not steal the frame from the main subject. - **Layout**: keep the main subject (the character and the jack-o'-lantern that is lighting up) exactly **in the center** of the frame, medium close shot, shallow depth of field, background street or porch blurred. - **Frame content**: the default fantasy scene "a jack-o'-lantern by the roadside suddenly lights up; the character and the lantern lock eyes" — the sudden warm light hits the character's face, frozen in the moment. Assemble the prompt as one paragraph in this order: - Opening: `Cinematic 3D Halloween scene: <one-line visual content>` - Style: `Style: festive Halloween atmosphere — jack-o'-lanterns, candlelight, drifting mist, bare branches and spider webs, warm orange firelight against a deep purple and blue night; volumetric light, soft rim light.` - Ghosts: `Ghost theme: only subtle ghost details — one or two translucent, faintly glowing spirit wisps drifting in the blurred background; they never dominate the frame.` - Layout: `Layout: main subject (the character and the jack-o'-lantern) centered in frame, medium shot, shallow depth of field, background street or porch soft and blurred.` - Camera: `Camera: slow push-in with a gentle handheld drift, cinematic 3D render, expressive characters.` - Dialogue (if any): `Dialogue (English): "<one line of English dialogue>"` — keep the dialogue in English and short. ### 2. Call `generate_video` - **With reference images**: `generate_video(prompt=..., input_urls=[lead character, second character, story location], model='ref2v', orientation='landscape', name='...', entity_type='final_videos')`. `input_urls` takes at least 1 and at most 3 images, in the order above; `ref2v` already handles character consistency, so do **not** also pass `gen_mode`. - **Text only**: `model='t2v'`, leave `input_urls` empty. - **User already has a still frame and wants it animated**: `model='i2v'`, put the still frame first in `input_urls`. - **Wants character consistency but gave no reference image**: first generate a base image with `generate_image(prompt=<description of the lead character>, model='t2i', name='Lead Character', entity_type='characte