popopo-99/zy-cinematic-realism

What it solves

This project provides a structured framework for transforming abstract story ideas, scripts, or reference images into precise, cinematic visual plans and prompts for AI image generators. It bridges the gap between narrative intent (who, where, and what is happening) and the technical prompt requirements of various AI models, preventing generic "AI-style" results and ensuring visual consistency across multiple frames.

How it works

It functions as a "Skill" for LLMs (like ChatGPT or Codex), acting as a visual director. Instead of simple prompt generation, it uses a systematic approach:

• Visual Development: It analyzes scripts to identify key narrative moments and develops "Scene Masters" (master scene descriptions) and visual rules.

• Director-Based Styling: It can apply the visual language of specific directors (from a library of 38) with adjustable intensity (slight, clear, or strong) without overriding the user's core scene facts.

• Dream Decoding: It analyzes reference images to extract specific patterns (color, exposure, material) rather than just copying the content, allowing those patterns to be applied to entirely new scenes.

• Model Compilation: It translates the finalized visual plan into optimized prompts for specific target models (e.g., Midjourney, Seedream, GPT Image) or provides a model-neutral version.

• Continuity Management: It generates "Continuity Bibles" and project handoff cards to maintain character and environment consistency across a series of images.

Who it’s for

• Storytellers and Screenwriters: To visualize scripts and develop art concepts for director discussions.

• AI Artists: Who want more cinematic control over lighting, camera angles, and composition rather than relying on random generation.

• Visual Concept Artists: Who need to maintain strict continuity across a sequence of frames.

Highlights

• Director Library: Access to a curated set of 38 cinematic styles to influence visual mood.
• Pattern Extraction: Ability to separate a reference image's style/lighting from its subject matter.
• Prompt Doctor: Tools for local repair and refining specific visual issues (e.g., removing "ad-like" lighting).
• Multi-Model Support: Adapters for various image generation models to ensure the prompt works across different platforms.
• Project Handoff: A system for recording state so a visual project can be resumed in a new chat session.