Piggyback Walk · عوامی Open Skill
Piggyback Walk
Piggyback Walk
Story
1 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/cqj5sgvf.jpg
# Piggyback Walk
Turn one reference image uploaded by the user into a short "carried on someone's back" video: the subject in the image is carried piggyback while moving forward. Single-shot, one-step generation, no structured state, and the output goes to the canvas automatically.
## When to use
- The user uploads (or points to) an image and asks for a short clip such as "carry them on the back", "piggyback walk", or "a video of them being carried while walking".
- The user is not satisfied with the generated result and asks to redo it, change the scene, change the lighting, or change the walking rhythm.
- The user only wants one thing changed (for example dusk instead of daylight, or a seaside path instead of the current background) — still use this skill, only edit the prompt, not the flow.
## Steps
1. **Confirm the input image**: there must be one reference image. Use the image the user uploaded in this turn, or a canvas image asset the user points to; prefer `asset_id` when referencing it, and `s3://` / https are also accepted. If there is no image at all, ask the user to upload a clear, fully visible subject with normal lighting first — do not generate one out of thin air.
2. **Read the subject**: decide from the image what the subject is, since it goes into the prompt later.
- If it is a person → that person is "the one being carried", and their identity features (face, hairstyle, clothing, colors, body proportions) must stay consistent.
- If it is a pet / doll / non-human subject → keep that subject's appearance unchanged, and let a person shown only from the back or in silhouette carry them while walking.
3. **Assemble one English prompt using the template in the next section**, replacing `{subject}` with the subject description from step 2.
4. **Call the generation tool**:
```
generate_video(
prompt=<the assembled English prompt>,
input_urls=[reference image reference],
model='i2v',
orientation=<see below>,
name='Piggyback Walk',
entity_type='final_videos',
)
```
- `model` is fixed to the channel `i2v` (image-to-video); never write a physical model name.
- `orientation`: follow the original image orientation — `'portrait'` for portrait / square images, `'landscape'` for landscape images; use `'portrait'` when the user explicitly asks for a vertical short clip.
- Do not pass `duration` / `resolution` by default (use the platform defaults); only pass them when the user explicitly asks for a specific length or clarity, and never pass an excessively long duration.
- Submit only one shot per call and wait for this tool call to return; do not resubmit the same task in the same turn, and do not split it into multiple shots or add other motion.
5. **Reply**: describe the generated clip in one or two natural-language sentences (who the subject is, the style and lighting, the feeling of the walk). Do not paste URLs to the user. If the tool returns `success=false`, state the failure reason honestly and never invent resource links. When the task is done and a next step should be suggested, call `set_guidance` in the same turn with options related to this clip (for example "Generate another one", "Try a different image", "Change it to dusk lighting").
## Prompt rules (this template must be used for the `generate_video` prompt)
The style, the scene content, and the layout must all make it into the prompt. Do not just write "generate a video of someone being carried on a back".
```
Cinematic short clip. {subject} is being carried on someone's back, both figures seen
from behind in a three-quarter back view, the carrier walks steadily forward with a
gentle rhythmic bounce and a natural stride, natural weight shift, slight follow-camera
tracking, motion smooth and continuous, the background slides past with parallax.
Style (one single unified visual tone for the whole frame): warm golden natural
daylight, soft rim light on the shoulders and hair, realistic skin an