AI Short Video Maker · সর্বজনীন Open Skill
AI Short Video Maker
Create ready-to-use short-form videos from a simple idea, text, product, image, or story. Generate an attention-grabbing hook, scene-by-scene visuals, short dialogue or voice-over, on-screen text, smooth transitions, and a clear ending. Keep the same characters, appearance, clothing, and visual style throughout the video when characters are used. Optimize videos for vertical 9:16 format and social media platforms including Instagram Reels, TikTok, Facebook Reels, and YouTube Shorts.
ShortDrama
6 3 6
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4203443/2026-10-03/6qjn2e8x.png
# AI Short Video Maker Turn a simple idea, text, image, product, or story into a polished 15–30 second vertical short video for Instagram Reels, TikTok, Facebook Reels, and YouTube Shorts. ## When to use - The user gives an idea, one-liner, text, product, or story and wants a short vertical video from it. - The user uploads an image (photo, product shot, character, scene) and wants a short video built around it. - The user wants to revise one stage of an existing short: rewrite the script, restyle a scene, regenerate a clip, redo the voice-over, or re-mix the final video. ## Reference files - `references/script-template.md`: The script structure (hook → scenes → CTA), scene/visual prompt conventions, and on-screen text rules. Read it before writing the script and before generating visuals. ## Steps 1. **Understand the input.** If the user uploaded an image, call `recognize_image(image_url=<asset URL>, prompt="Describe the main subject, setting, style, and any text or product details so I can build a short video around it.")`. If the input is text/idea/story, parse the core message, target audience, and desired tone. Clarify with the user only if the intent is genuinely ambiguous (e.g. no clear subject or goal). 2. **Write the script.** Read `references/script-template.md`, then produce a 15–30 second vertical script: a strong hook (first 2–3 seconds), 3–6 scenes, dialogue or voice-over lines, on-screen text per scene, and a clear call-to-action when appropriate. Save it with `create_document(content=<script>, name="<title> script", entity_type="scripts")`. Show the user the script and confirm before generating assets. 3. **Generate scene visuals.** For each scene, call `generate_image` with `orientation='portrait'`, `resolution='720P'`, `model='t2i'` (text-only input) or `model='i2i'` with the uploaded image in `input_urls` (image-based input). Keep characters consistent by reusing the same character description in every scene prompt. Pass `name="scene_<n>_<short title>"` and `entity_type="storyboard_images"`. 4. **Generate the voice-over.** Call `generate_tts(text=<dialogue/VO lines>, model='tts.design', voice_instruction=<English voice description matching the tone, e.g. "energetic, upbeat female narrator in her 30s">)`. Pass `name="<title> voiceover"` and `entity_type="voiceovers"`. 5. **Generate background music.** Call `generate_music(prompt=<style matching the mood, e.g. "upbeat modern pop, driving beat, suitable for a 20-second social media reel">, duration=20)`. Pass `name="<title> music"` and `entity_type="music"`. 6. **Generate scene clips.** For each scene image, call `generate_video(input_urls=[<scene image asset_id>], model='ref2v', prompt=<motion description for that scene>, orientation='portrait', duration=<2–5s per scene>). Use `ref2v` to preserve character/subject consistency across scenes. Pass `name="clip_<n>_<short title>"` and `entity_type="storyboard_videos"`. 7. **Mix the final video.** Call `mix_clips` with one clip per scene: `video=<scene clip>`, `audio=<voiceover>`, and add the music as a second audio track (`audios`). Burn in the on-screen text with `subtitles.items` (each cue's `text`/`start`/`end` relative to the clip). Use `orientation='portrait'`. Pass `name="<title> final"` and `entity_type="final_videos"`. 8. **Deliver.** Summarize the final video in natural language (hook, scenes, VO, music, CTA). Do not paste raw URLs. Offer next-step tweaks (rewrite script, restyle a scene, regenerate a clip, redo VO, re-mix). ## Notes - Always use `orientation='portrait'` (9:16) for vertical short-form video. - Keep characters consistent: reuse the exact same character description across all scene prompts and use `ref2v` for clips. - On-screen text should be short (a few words per scene), readable, and match the spoken line. - Report failures honestly; never fabricate URLs. Politely decline requests for prohibited content (video generation does not support NSFW/adult content on any channel). -