Kissing Video Generator · öffentlicher Open Skill
Kissing Video Generator
Kissing Video Generator ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.
Other
1 0 3
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284863/2026-10-20/qsf9ahvp.jpg
--- allowed-tools: - recognize_image - generate_image - generate_video --- # Kissing Video Generator Turn user-uploaded reference photos into one short cinematic kissing video that keeps the same people, faces, wardrobe and mood as the photos. This is a reference-photo-to-short-video app: the photos are the source of identity and styling, and the output is a single short clip, not a still image and not a multi-scene film. There is no structured project state. Keep the plan in the conversation, use `write_todos` for multi-step runs, and rely on the canvas assets produced by the generation tools. ## When to use - The user uploads one or more photos and asks for a kissing video, a kiss clip, a romantic couple video, or "make them kiss". - The user uploads a man photo and a woman photo and wants them brought together in a short video. - The user asks to change the mood, framing, duration or lighting of a kissing video that was already generated in this conversation. - Do not use for plain still-image requests (use image generation only) or for unrelated video editing of existing footage. ## Steps 1. **Check the inputs.** Read `references/step-details.md` and follow its input branches. You need at least one clear reference photo with visible faces. If the user gave only one person, ask for the partner photo before generating; do not invent a partner unless the user explicitly allows it. 2. **Plan the run.** For anything beyond a single generation, call `write_todos` with the short list of steps (analyze, compose, script, generate, review) and keep exactly one item `in_progress`. 3. **Analyze the reference photo(s).** Call `recognize_image` on each uploaded photo with a prompt asking for: number of people and their left/right positions, gender presentation, face and hair details, facial hair, clothing colors and materials, accessories, setting, time of day, light direction, background elements and current framing. Record the result in the conversation; it is the identity lock for every later prompt. 4. **Optional composition still.** If the two people come from separate photos, or the framing needs to be unified, create a single composed still first with `generate_image(model='i2i.hd.lite', input_urls=[...both reference photos...], orientation='portrait', name='Kissing composition', entity_type='storyboard_images')`. Build that prompt from `references/style-and-prompt.md` (read it first in step 5). Treat the still as the visual anchor, not as the final product. 5. **Read the style and prompt rules.** Call `read_skill_file(path="references/style-and-prompt.md")`. Use its Style, Subject/content, Layout, Must keep, Must avoid and prompt templates to write the English prompt for this specific request. Never fall back to a generic template without filling in the analyzed photo details. 6. **Write the video script before generating.** Call `read_skill_file(path="references/video-script.md")` and write a complete English script for this request: logline, duration, cast and identity locks, setting, beat-by-beat breakdown with shot size, camera move, subject action and light for each beat, plus must-keep and must-avoid. The script must follow the current user intent (mood, pacing, framing preference, duration) and the actual photo analysis; a generic template is not an acceptable script. 7. **Read the video constraints.** Call `read_skill_file(path="references/video-content-constraints.md")` and apply its output specs, shot grammar, motion limits, reference-video input rules and forbidden content to the script and the prompt. 8. **Generate the video.** Call `generate_video` with: - `prompt`: the compressed English prompt built from the script and the style rules (style, both subjects with identity locks, action beats, camera move, lighting, mood, exclusions). - `input_urls`: the uploaded reference photo URL(s); for a two-photo input pass both, man first then woman. If a composition still was made, pass it as the first inp