Halloween Costume Video Generator · skill Open Skill công khai
Halloween Costume Video Generator
Halloween Costume Video Generator
Manhua
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/hvakapsc.jpg
# Halloween Costume Video Generator Turns an image the user uploaded into a Halloween costume version: first produce a "costume hero image" (new costume + Halloween scene / props / lighting, while preserving the subject's identity and the original framing), then generate a short Halloween costume video based on that still. There is no structured project state; cross-turn reuse relies only on assets that already exist on the canvas. ## When to use - The user uploads or references an image and says "Halloween costume", "put them in a Halloween outfit", "make a Halloween-themed image / video". - Only the costume is changed (keep the original background), only the background is changed (keep the original costume), or both are changed. - The user wants just one costume hero image, or wants a costume-transformation video clip. - The user asks to redo an already generated costume image / video: different look, different scene, longer duration. ## Steps ### 1. Get the reference image and the look intent - **The reference image is a required input for this skill**: an image asset attached to the user message, or an image already on the canvas — pass its `asset_id`. If there is no image, ask the user to upload one first; never invent a person from nothing. - When you need to actually see the subject, call `recognize_image(image_url=<image asset_id>)` to work out who / what it is, what they are wearing, the pose and the composition direction, and write the prompt from that. Do not guess the content from the file name. - Look: if the user names one, follow it; if they only say "surprise me", pick the one from the "Look library" below that best matches the subject's vibe, and tell them in your reply which one you picked. - Orientation, duration and other defaults follow the conventions below — do not ask the user about them. ### 2. Generate the costume hero image (new costume + Halloween scene) ``` generate_image( prompt=<English prompt filled from the "Prompt conventions" templates below>, input_urls=[<reference image asset_id>], model='i2i', # image-to-image. If the user explicitly wants higher quality AND the look has no realistic gore, you may switch to 'i2i.hd' orientation=<match the original image: 'portrait' for vertical portraits, 'landscape' for horizontal, 'square' for square>, name='Halloween Costume - <look name>', entity_type='characters' ) ``` Build the prompt in six parts — identity lock → costume → scene and props → lighting → framing and layout → style closer — all written in English. ### 3. Generate the costume transformation video Use the still from step 2 as the first frame: ``` generate_video( prompt=<English prompt filled from the video template below>, input_urls=[<step 2 still asset_id>], model='i2v', orientation=<same as the still; for a square original, pick 'portrait' or 'landscape' based on which way the subject faces>, duration=5, name='Halloween Costume - <look name>', entity_type='final_videos' ) ``` - The video prompt should describe only the "bring it to life" part (fabric, fog, candle flames, jack-o'-lantern light, bats, a small subject movement, camera motion); the costume, identity and background must stay identical to the still — do not describe a different look again. - If the user explicitly says "image only" → skip this step and reply directly. - Before starting step 2, use `write_todos` to list the 2 steps (costume hero image / costume video), and update the whole list's `status` as you progress. ### 4. Reply In one or two natural-language sentences, say what you did (which look was applied, which Halloween elements were added, how long the video is). Do not paste long URLs. If the task is done, call `set_guidance` in the **same turn** with relevant next steps, e.g. "Do another version as a witch", "Extend the video to 10 seconds", "Change only the background to a graveyard, keep the costume". ## Prompt conventions ### Style - One-line tone: **cinem