prompts.

胡金铨风格的客栈武打，竖屏10秒

总裁把甩出黑卡，三分讥笑七分凉薄。10秒，竖屏短剧质感

**SKILL-Pro (专业外挂版)**：搭载 7300+ 行庞大知识库。引入 **RAG (检索增强) 动态加载机制**，仅在用户需要深究王家卫风格、希区柯克运镜等极客需求时加载，拒绝上下文乱飞。

"胡金铨风格的客栈武打，竖屏10秒"

A serene mountain landscape at sunset with golden light reflecting on a peaceful lake

A futuristic city floating in the clouds

A cat wearing a astronaut helmet, digital art, trending on artstation

Change the sky to a starry night

This repository contains a Cog implementation of **BAGEL (Unified Model for Multimodal Understanding and Generation)**, ByteDance's incredible 7B parameter multimodal model that can generate images, edit images, AND understand images—all in one unified model. This is the kind of AI we've been waiting for: one model that truly does it all.

/nanobananas-api-guide How do I generate an image with Python? /nanobananas-api-guide What are the video generation pricing options? /nanobananas-api-guide Help me set up async image generation in Node.js

A comprehensive Claude Code skill for building with the [NanoBananas.AI](https://nanobananas.ai) API. Covers image generation, video generation, pricing, authentication, and includes ready-to-use code examples.

This project offers a collection of ready-made prompts and templates for AI video generation using Seedance 2.0. It helps you create high-quality videos in styles like cinematic film, anime, user-generated content (UGC), social media clips, memes, and ads.

You do not need programming skills to use these resources. The templates guide the AI to generate videos with great detail and style.

A modern graphics card with DirectX 12 support for better video rendering (optional but recommended)

Make sure your PC meets these requirements for a smooth experience.

Use provided templates for various genres like cinematic, anime, memes, ads, or social clips.

Generate a UGC selfie of Sofia holding the cola can in her bedroom

Generate a UGC video ad for this product

Generate 10 thumbnail variants for this video about prompt engineering

The agent generates a full-body hero image for your approval via Nano Banana 2, then creates 9 additional angles (3/4 views, profile, closeup, etc.) using the hero as a reference URL. All 10 images are saved to `references/influencers/` for future use.

分镜规划 → 图像生成 → 视频合成 → 音乐制作 → 音视频对齐 → 多平台导出。

上传音频后，生成前先完成结构分析，驱动分镜时长规划：

**段落结构分割** — MFCC + chroma 特征 → 递归相似矩阵 → agglomerative 聚类 → 按 RMS 能量分配 intro / verse / chorus / bridge / outro 标签

Frame Chaining**：上一镜头末帧作为下一镜头参考图，保持视觉连贯性。