Squat Video Generator · öffentlicher Open Skill
Squat Video Generator
Squat Video Generator ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.
Trailer
9 0 5
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-20/ubyctv8e.jpg
# Squat Video Generator Turn one reference photo into a short vertical video in which the same person performs continuous, controlled bodyweight squats. The uploaded person's identity is locked; wardrobe, posture, and set are restyled for the squat theme. Optional text from the user controls mood, wardrobe, set, tempo, and duration. ## When to use - The user uploads a photo and asks for a squat video, workout clip, fitness motion video, or "make this person squat". - The user uploads a photo plus text such as "gym, black sportswear, 10 seconds" and wants a short video. - The user asks to revise an already generated squat video (tempo, clothing, background, duration, framing). - Do not use for still-image-only requests; skip the script and video steps in that case. ## Inputs and outputs - Input: one reference image (preferred) and optional text instructions. - Output: one short vertical video, single continuous shot, identity preserved. - The reference image supplies the person's face, hair, skin tone, body proportions, and any distinctive features. - The user's text supplies wardrobe, set, lighting mood, squat depth, tempo, number of repetitions, and duration. Never ignore explicit text. - If no image is supplied but the user describes a person, create a first frame with `generate_image` using `model='t2i'`, then animate it. ## Steps 1. Read `references/workflow-stages.md` to get the intake checklist, stage order, channel choices, branch rules, and retry limits. For multi-step work, call `write_todos` with one item per stage. 2. Before any generation, call `read_skill_file` for `references/style-and-prompt.md` and `references/video-content-constraints.md`. Follow their style, framing, must-keep, must-avoid, and motion rules exactly. 3. Inspect the reference image with `recognize_image`. Ask it to describe the person's identity features, current clothing, pose, framing, background, and image quality. Use the result to fill the identity-lock and first-frame fields. If the image is unusable, follow the branch rules in `references/workflow-stages.md`. 4. Prepare the first frame: - Uploaded image is usable: use it directly as the first frame. - Uploaded image needs reframing or a squat-ready starting pose: call `generate_image` with `model='i2i.hd.lite'`, `input_urls=[reference image]`, `orientation='portrait'`, plus the image prompt from `references/style-and-prompt.md`. Keep the face and body identity unchanged. - No image at all: call `generate_image` with `model='t2i'`, `orientation='portrait'`, and the text-to-image prompt from `references/style-and-prompt.md`, then use the result as the first frame. - Use a short `name` and `entity_type='storyboard_images'` for prepared frames; use `entity_type='characters'` only for character reference sheets. 5. Read `references/video-script.md` with `read_skill_file`. Write a complete English script for this specific request: logline, duration, subject lock, wardrobe and set, motion beats with timing, camera, lighting, must-keep, must-avoid. Fill it from the user's exact text and the image analysis. Never reuse a generic template as the finished script. 6. Compress that script into one English video prompt using the compression order in `references/video-script.md` and the templates in `references/style-and-prompt.md`. The prompt must describe motion over time, not just restate the still image. 7. Call `generate_video` with `model='i2v'`, `input_urls=[first frame]`, `orientation='portrait'`, `duration=5` by default (use 10 only when the user asks for a longer clip), a short `name`, and `entity_type='final_videos'`. Do not pass a physical model name. Do not add a last frame. 8. Check the returned result. If `success=false`, report the real error and retry at most twice following the retry rules in `references/workflow-stages.md`. On success, reply in one or two natural sentences describing the clip; do not paste long URLs. In the same turn, call `set_guidance` with releva