Private Swim Coach Video · Open Skill 公开技能
Private Swim Coach Video
Private Swim Coach Video 是 Digen 上的公开 Open Skill。打开后即可了解该工作流能制作什么。
Roleplay
0 0 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-21/bx2xeqfb.jpg
# Private Swim Coach Video
Turns one uploaded character photo into a short vertical video in which the same person gives a coy, camera-flash close-up look and silently asks for "a private swimming coach" with her expression. Everything — palette, flash-lit skin, teal industrial ceiling, framing, gaze — stays inside that single visual world. There is no structured project state; the uploaded photo plus the assets on the canvas are the only memory.
This skill is a character-reference app: identity, hair, makeup and wardrobe come from the user's photo, never from invention.
## When to use
- The user uploads a photo (selfie, portrait, character art) and wants a short video of that person in this app's style.
- The user asks for "another take", "make it longer / shorter", "different expression", or a re-render of a clip this skill already produced.
- The user describes the idea loosely ("she asks me to be her private swimming coach", "flirty look up at the camera") while attaching a reference image.
Do not use it for pure text-to-image requests, for multi-shot edited films, or when there is no reference image at all (ask for one first — identity must come from a real upload).
## Steps
1. **Inspect the uploaded reference photo.**
Call `recognize_image` with `image_url` set to the uploaded image (and a short prompt asking for face, hair colour, makeup, wardrobe, necklace, background, lighting and framing). Use the answer to fill the identity slots in the prompt templates. If the tool fails, continue with what is visibly obvious and say nothing about the failure unless it changes the result.
2. **Read the visual rules before generating anything.**
Call `read_skill_file(name="private-swim-coach-video", path="references/style-and-prompt.md")`. It contains Style, Subject/content, Layout, Must keep, Must avoid and the fill-in prompt templates. Every image and video prompt in this skill must be written against that file — do not improvise your own look.
3. **Build the keyframe still (image-to-image).**
Fill the keyframe template from `references/style-and-prompt.md` with the identity details from step 1 and write the prompt in English. Then call:
`generate_image(prompt=<filled keyframe template>, input_urls=[<uploaded photo>], model='i2i.hd.lite', orientation='portrait', name='Swim Coach Keyframe', entity_type='storyboard_images')`
The uploaded photo is the identity and wardrobe source; the template supplies gaze, expression, flash lighting, palette and the teal industrial ceiling. Keep the still vertical — this app is always portrait.
4. **Read the motion limits.**
Call `read_skill_file(name="private-swim-coach-video", path="references/video-content-constraints.md")` for the allowed shot size, camera behaviour, motion amount, duration feel, expression-mimicry boundaries and the list of forbidden content. These limits override anything the user asks for that would break the app's look.
5. **Write this request's own script (mandatory before any video call).**
Call `read_skill_file(name="private-swim-coach-video", path="references/video-script.md")`. Following that structure, write a complete English script for *this* user request — logline, duration, beat-by-beat shots with shot size, camera move, subject action, lighting/atmosphere, transitions, must-keep and must-avoid — using the user's wording and the step-1 image analysis. Do not paste a generic template as the finished script; if the user asked for a different mood or length, the beats must change accordingly.
6. **Compress the script into the video prompt and generate.**
Condense the script into one continuous English prompt (script beats in order, not a bullet list of abstractions), then call:
`generate_video(prompt=<condensed script>, input_urls=[<uploaded photo>, <keyframe from step 3>], model='ref2v', orientation='portrait', duration=5, name='Private Swim Coach', entity_type='final_videos')`
`ref2v` keeps the face consistent; the keyf