Dachshund-Skater — Open Skill
Dachshund-Skater
Dachshund-Skater is a public Open Skill on Digen. Open it to see what this workflow creates.
Sports
11 1 4
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/1856918/2026-09-21/fq7u22u4.png
# Dachshund Skater Turn an uploaded photo of a person into a playful, cinematic video where that person rides on a fast-running dachshund that acts as their skateboard. The uploaded person is the identity anchor; the dachshund is a new element described in the prompt (never taken from the reference). ## When to use - User uploads a person's photo and wants a "dachshund skateboard" video (standing on the dog's back, doing tricks). - User wants to change the setting, add tricks, or extend the clip. ## Key constraints - **Channel:** always `generate_video(model='ref2v')` with the uploaded person's image in `input_urls`. ref2v preserves the person's identity. - **Duration clamp:** ref2v caps each single render at ~10 seconds regardless of the `duration` value. For a longer clip, render multiple 10s segments and stitch them with `mix_clips`. - **Orientation (default 9:16):** by default render in **9:16 (portrait)** — do not ask the user for the format. Only use a different format (e.g. 16:9 landscape) if the user explicitly requested it beforehand. Use `orientation='portrait'` for 9:16 and `'landscape'` for 16:9. - **Fisheye filter:** ask whether the user wants a **fisheye camera filter**. If yes, add an ultra-wide fisheye lens (strong barrel distortion, exaggerated perspective, curved edges) to the camera direction in the prompt; if no, keep a normal lens. - **Framing (mandatory):** the dachshund is the main focus of every shot — the camera should center on the dachshund and its tricks. The person from the uploaded photo only needs to be briefly visible in between; otherwise only their legs are visible (standing on the dog's back). Use gentle, following camera movements that smoothly track the ride — no dramatic pans, zooms, whip pans, or fast tracking shots. - **Dachshund consistency (mandatory):** the dachshund is always the same fixed character — a long-bodied, short-legged, chestnut-brown dachshund with short fur. Keep this exact description identical in every prompt and across all segments of a longer video so the dog never changes color, size, or fur between clips. - **Resolution:** always `resolution='720P'`. - **Prompt language:** English for the ref2v prompt structure. - **Audio note:** ref2v does not guarantee audio. The generated video may be silent; if the user wants sound, generate/mix it separately (e.g. via `mix_clips` with a music/ambience track) rather than expecting it from the render. - **The dachshund** is a long-bodied, short-legged, chestnut-brown dog with short fur that sprints very fast and behaves like a skateboard. It flips in the air for flip tricks (kickflip, heelflip, 360 flip), grabs its own body in the air for grab tricks (indy, melon, nose, method), slides on its front paws or side for grinds (nose-slide, 50-50, boardslide), rears onto its hind legs for manuals (manual/wheelie, nose manual), and leaps for an ollie. ## Steps 1. **Ask filter (if not specified).** The format defaults to **9:16 (portrait)** — do not ask. Only use another format if the user already stated it. Ask whether they want a **fisheye camera filter** (yes/no); skip if already answered. 2. **Lock identity.** Call `recognize_image` on the uploaded person's image to capture face, hair, body type, clothing, colors, and distinctive features. Use these in the prompt so the person stays recognizable. If the image does not show a clear single person (e.g. a group photo or no person), ask the user which person is the right one before continuing. 3. **Compose the ref2v prompt.** Use the six-part template in `references/prompt-template.md`. Fill in the person's identity from step 2, the requested setting/tricks, and the fisheye filter. Keep the person's facial expression natural and believable (focused during tricks, satisfied on landing) — never stiff or blank. In every shot, keep the dachshund as the main focus; the person from the uploaded photo only needs to be briefly visible in between, otherwise only their legs are visible. Use g