Girl Playing Football — AI Video Same-Style · skill pública de Open Skill
Girl Playing Football — AI Video Same-Style
Girl Playing Football — AI Video Same-Style es una Open Skill pública en Digen. Ábrela para ver qué crea este flujo.
Sports
2 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-16/3ptbw54t.jpg
# Girl Playing Football — AI Video Same-Style Turn the user's uploaded reference image into a short video: the subject of the uploaded image is re-staged as a girl playing football, with the tuned camera rhythm and picture texture of the original "AI video same-style" trend clip. There is no structured project state — the inputs are the user's image(s) plus their message, and the outputs are canvas video assets (plus, optionally, one prepared keyframe image). ## When to use - The user uploads one or more images and asks for a short video, "make it move", "animate this", or "same-style as this AI video". - The user asks for the girl-playing-football trend clip, or asks for a re-run with a different reference image. - The user wants to change only one part of an already generated clip (camera, motion, duration, or a detail that must be kept). Redo only that clip — do not restart from scratch. ## Steps 1. **Read the look book before generating anything.** Call `read_skill_file(name='girl-playing-football-video', path='references/style-and-prompt.md')` and follow its Style / Subject / Layout / Must-keep / Must-avoid rules. Read `references/video-content-constraints.md` before the video call, and `references/workflow-stages.md` when you need the sub-step checklists, branches and retry rules. Do not read all three at once — read the file for the step you are on. 2. **Lock the input.** Identify the reference image the user attached (asset id or URL). Exactly one image is the anchor. If the user attached several, ask which one is the subject to keep; if the user attached none, ask for one clear image before generating anything. 3. **Read the reference image.** Call `recognize_image(image_url=<reference image>, prompt='Describe the main subject, its identity cues, clothing and composition')`. From the answer, write a short English must-keep list (face/identity cues, hair, clothing colours, distinctive accessories or props) and use it verbatim in every prompt afterwards. 4. **Optional keyframe prep (only if needed).** If the uploaded image is low resolution, cluttered, off-centre, or not in the target aspect ratio, prepare a clean keyframe first: `generate_image(prompt=<i2i prompt from references/style-and-prompt.md, begin with the must-keep list>, input_urls=[<reference image>], model='i2i.hd.lite', orientation=<see step 5>, resolution='720P', name=<short subject name>, entity_type='storyboard_images')`. Keep the same person and the same clothing colours; only re-frame the subject on a sunlit football pitch. If the uploaded image is already clean, skip this step and use the original image. 5. **Generate the clip.** Call `generate_video(prompt=<composed from references/style-and-prompt.md + references/video-content-constraints.md>, input_urls=[<keyframe from step 4 or the original reference image>], model='ref2v', orientation=<portrait by default; match the reference image when it is clearly landscape>, resolution='720P', duration=5, name=<short clip name>, entity_type='final_videos')`. - `ref2v` is reference-to-video: it keeps the subject's identity. Do **not** pass `gen_mode` — the channel already decides the mode — and never pass a physical model name. - Prompt assembly order: must-keep list → the girl-playing-football re-staging → the action beats → the tracking camera → the unified look → the avoid list. The copy-paste fragments live in `references/style-and-prompt.md`. - Default `orientation='portrait'` and `resolution='720P'`; use `1080P` only when the user asks for higher quality. Do not pass `width`/`height`. Leave out `duration` beyond the default unless the user asks for a different length. 6. **First/last-frame branch (fallback only).** If the user supplies (or approves) an explicit start frame and end frame, or the reference-to-video run cannot hold the action arc after two retries, call `generate_video(prompt=..., input_urls=[<start frame>, <end frame>], model='i2v', orientation=<same as above>, resolution='7