SnapLeap Jump Shot · skill pública Open Skill
SnapLeap Jump Shot
SnapLeap Jump Shot é uma Open Skill pública no Digen. Abra para ver o que este fluxo cria.
Custom
1 0 2
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284862/2026-10-19/spqu6s9j.jpg
# SnapLeap Jump Shot Turn one uploaded portrait into a single bright outdoor action photo: the same person, caught mid-air in a joyful leap, shot from a low wide-angle viewpoint against a vivid blue sky. One input photo, one output image, identity preserved. ## When to use - The user uploads a photo and asks for a "jump shot", "leaping photo", "dynamic outdoor shot", "action selfie", "mid-air photo". - The user wants their portrait re-shot in a sunny outdoor scene with an energetic pose, keeping their own face. - The user asks to redo a previous result with a different pose, outfit or outdoor scene. - Do not use for free illustration/poster design, group photos, or anything requiring text on the image. ## Inputs and outputs - Input: one reference photo of a single person (required). Use it as `input_urls` for image-to-image generation; it is both identity reference and the source of framing. - Output: one image asset on the canvas. No video, no multi-panel poster, no text overlay. - Aspect/quality defaults: `orientation='portrait'`, `resolution='1080P'`. Only change orientation if the user explicitly asks for a landscape crop. ## Steps 1. **Check the input.** Confirm exactly one clear, well-lit photo of one person, face visible and not heavily cropped. If no photo was uploaded, ask for one — this skill always starts from a photo. If the photo is blurry, has several people, or the face is tiny/obscured, say what is missing and ask for a clearer one; only proceed with a best-effort attempt if the user insists. 2. **Inspect the photo when in doubt.** Call `recognize_image(image_url=<user photo>, prompt="Describe the person's face, hair, apparent age, current outfit and the framing; note how sharp the face is.")` when you need to confirm a single subject, framing or wardrobe before writing the prompt. 3. **Read the visual rules.** Call `read_skill_file` for `references/style-and-prompt.md` (style, subject, layout, must-keep, must-avoid, prompt template) and `references/scene-and-pose-library.md` (pose + outdoor scene options). Read `references/step-details.md` when the input needs a quality decision or a regeneration. 4. **Lock identity, then choose pose and scene.** Keep the uploaded person's face, hairstyle and skin tone; change only wardrobe, pose and location. Pick one pose and one scene from the library (wheat field leap is the default) unless the user named a scene. 5. **Write the English prompt.** Assemble it with the template in `references/style-and-prompt.md`: identity lock first, then pose, then low-angle wide-angle composition, sky/foreground, lighting, colour, texture, mood, and the must-avoid tail. Never ship the template unfilled — every bracket must be replaced with concrete content taken from the user's photo and the chosen scene. 6. **Generate.** `generate_image(prompt=<filled English prompt>, model='i2i.hd.lite', input_urls=[<user's uploaded photo>], orientation='portrait', resolution='1080P', name='Jump Shot', entity_type='characters')` - `model` is always the `i2i.hd.lite` channel; never pass a physical model name. - Keep the user's photo as the first (identity) input; do not substitute a stock or library image for their face. 7. **Self-check before delivering.** Run the checklist in `references/step-details.md`: same person, one subject, correct camera angle, no text/poster elements, believable anatomy. If identity is broken or limbs/hands are malformed, regenerate once with a tightened identity sentence and simpler pose. 8. **Reply and suggest next steps.** Two or three natural sentences about what changed (pose, scene, light). Do not paste long URLs. In the same turn, call `set_guidance` with concrete follow-ups such as another scene, a different pose, or a landscape crop. ## Notes - Identity is the contract: the face must stay recognisably the uploaded person. Never describe a different face, age or ethnicity, and never merge two people. - The output is a photograph, not a promotional pos