Cute Girlfriend Video · publiczny skill Open Skill
Cute Girlfriend Video
Cute Girlfriend Video to publiczny Open Skill w Digen. Otwórz go, by zobaczyć, co tworzy ten workflow.
TryOn
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/sx28pvfq.jpg
# Cute Girlfriend Video
Turn one uploaded photo into a short, warm, cozy "cute girlfriend" clip. The face, hair and styling of the person in the upload must survive the whole pipeline: photo analysis -> identity-preserving keyframe -> scripted single-shot video.
There is no structured project state. Keep the identity anchors (hair, outfit, palette, framing) in your own notes / the prompt, and reference already generated assets by their URLs.
## When to use
- User uploads a photo (selfie, portrait, half-body shot) and asks for a short video of "my cute girlfriend", "make this photo move", "turn my photo into a video".
- User wants a variant of an already produced clip: different gesture, different room mood, longer/shorter, softer or more playful.
- Do **not** use for pure still-image requests (skip the video steps) or for clips that need multiple cuts/subtitles/editing of existing footage.
## Steps
1. **Check the input.** Exactly one clear adult-person photo is the ideal input; several are fine (use the clearest face as the identity anchor). If no image was uploaded, ask for one before doing anything else.
2. **Read the working checklist, then analyze the photo.**
- `read_skill_file(name="cute-girlfriend-video", path="references/step-details.md")` and follow its "Step 1" checklist.
- Call `recognize_image(image_url=<uploaded image>, prompt=<English prompt asking for: face/identity features, hair length-color-style, makeup, outfit pieces and colors, pose and camera angle, background and props, light sources and color temperature, overall palette>)`.
- Keep the returned details as the identity anchor for every later prompt.
3. **Read the visual rules, then build the keyframe.**
- `read_skill_file(name="cute-girlfriend-video", path="references/style-and-prompt.md")` and use its Style / Subject / Layout / Must keep / Must avoid sections plus the keyframe prompt template to compose a fresh English prompt for this user (never paste the template verbatim).
- Call `generate_image(prompt=<composed English prompt>, input_urls=[<uploaded reference image>], model='i2i.hd.lite', orientation='portrait', name='Girlfriend Keyframe', entity_type='storyboard_images')`.
- If the result loses the face or the outfit, follow the retry rules in `references/step-details.md` before continuing.
4. **Write this run's video script.**
- `read_skill_file(name="cute-girlfriend-video", path="references/video-script.md")` and fill every section for the **current** request: logline, duration, beats with shot size / camera move / subject action / light / transition, continuity anchors, must-keep, must-avoid.
- Derive it from the user's own words (mood, gesture, setting, length) plus the photo analysis and the keyframe you just made. A generic template is not a script: each beat must name concrete action and camera behaviour for this person in this room.
5. **Read the video limits, then compress the script into the video prompt.**
- `read_skill_file(name="cute-girlfriend-video", path="references/video-content-constraints.md")` and apply its shot / motion / duration / identity constraints.
- Compress the script into one English prompt (identity anchor -> keyframe look -> ordered action beats -> camera -> light and mood -> hard constraints), then call
`generate_video(prompt=<compressed English prompt>, input_urls=[<keyframe image>, <uploaded reference image>], model='ref2v', orientation='portrait', duration=<5 by default>, name='Cute Girlfriend Clip', entity_type='final_videos')`.
- Use the branch table in `references/step-details.md` if the user asks for a strictly still-anchored opening shot (then `model='i2v'` with the keyframe as the first frame) or for extra moments (repeat steps 4-5 for each additional clip).
6. **Reply.** One or two sentences on what was produced (who it shows, the mood, the length), never long URLs. In the same turn call `set_guidance` with 2-3 next actions that fit this clip (e.g. "Make it