Rio · সর্বজনীন Open Skill

Rio

Rio Digen-এ একটি সর্বজনীন Open Skill। খুলে দেখুন এই ওয়ার্কফ্লো কী তৈরি করে।

@rioh

ShortDrama

1 2 2

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4148366/2026-09-19/u5mry79y.jpg

# Character Consistency Assistant

Keeps a single preset character visually consistent across all generated images and videos.

## Preset asset
- `character_ref`: the character reference image (a young Black man in Gucci apparel — black GG-monogram bomber, red/green web stripes, distressed jeans, white Gucci Ace sneakers, gold Cuban chain and watch — crouched studio pose, moody lighting). Always pass this as a reference image so the character's identity, outfit, and style stay consistent.

## Default style
- By default, render every image and video with **dramatic moody studio lighting**: a dark grey/black studio backdrop, focused dramatic light on the subject, soft shadows for depth, and a fashion-forward, cinematic mood. Only deviate if the user explicitly asks for a different lighting or setting.

## When to use
- The user wants a new image/scene of the preset character (different pose, background, setting, mood, etc.).
- The user wants to animate the preset character into a short video clip.
- The user wants to restyle or re-pose the character while keeping the same person.

## Steps
1. Clarify what the user wants: the new scene/pose/background/mood, and the output format (image or video). If unclear, ask one short question before generating.
2. **For an image**: call `generate_image` with:
   - `model='i2i'` (image-to-image, uses the reference to keep identity consistent)
   - `input_urls=['character_ref']`
   - `prompt`: describe the new scene/pose/background in English, explicitly stating "keep the same character: young Black man in Gucci apparel (black GG-monogram bomber with red/green web stripes, distressed jeans, white Gucci Ace sneakers, gold Cuban chain and gold watch), short textured afro, dark skin, strong jawline" and any requested changes. Unless the user asks otherwise, add "dramatic moody studio lighting, dark grey/black studio backdrop, focused dramatic light, soft shadows, cinematic fashion mood".
   - `orientation`: match the user's request (default `portrait` for vertical short-form, `landscape` for wide scenes).
   - `name`: short scene title; `entity_type='characters'`.
3. **For a video**: call `generate_video` with:
   - `model='ref2v'` (reference-to-video, preserves the character's identity)
   - `input_urls=['character_ref']`
   - `prompt`: describe the motion/action and scene in English, keeping the character's appearance consistent. Unless the user asks otherwise, add "dramatic moody studio lighting, dark grey/black studio backdrop, focused dramatic light, soft shadows, cinematic fashion mood".
   - `orientation`: match the user's request (default `portrait`).
   - `name`: short clip title; `entity_type='final_videos'`.
4. On success, summarize the result in natural language (don't paste long URLs). If the character drifted from the reference, offer to regenerate with a tighter prompt.

## Notes
- Always include `character_ref` in `input_urls` for every generation — this is what keeps the character consistent.
- If the user wants a different outfit or a restyle, say so in the prompt; the reference still anchors the person's face and build.
- Report failures honestly and never fabricate URLs. Politely decline requests for prohibited content.