Creative Intent Format (CIF) — A lightweight standard for preserving creator identity in generative AI

[Tricksterpaws]

Hi WICG team,

I’m an independent digital creator, not affiliated with any company. I’ve been using AI image tools and hit a wall: every time I generate a new image of my character, I lose something essential unless I retype it all.

That led me to propose Creative Intent Format (CIF) — a lightweight, user-owned standard to preserve sacred traits across generations.

I’ve published a draft spec here:
👉 https://github.com/Tricksterpaws/Creators-Bridge

I’d love your feedback — could this be a web-native capability for creators?

—Tricky (Ethan)

---

Explainer: Creative Intent Format (CIF) — A Web Standard for Preserving Creator Intent in Generative Media

Author: Tricky (Ethan), independent digital creator
Status: Draft — Seeking community feedback
License: CC0 1.0 Universal

🎯 Problem: Generative AI Lacks Fidelity to Creator Intent

Web-based generative AI tools (text-to-image, text-to-3D, etc.) are increasingly used for creative expression, storytelling, and design. However, users cannot reliably preserve core visual or conceptual traits across generations.

For example:

“Draw my character again, but in space—keep her silver mask and twin scars.”

Current systems treat each prompt as stateless, forcing creators to manually re-specify “sacred” traits every time. This leads to:

• Visual drift (loss of identity)
• High cognitive load (repetitive prompting)
• Fragmented creative workflows (no persistent character/scene memory)

While models support weighting (e.g., (mask:1.3)), there’s no standard way for web apps to capture, store, or transmit structured creative intent across sessions or tools.

💡 Proposal: Creative Intent Format (CIF)

I propose CIF—a lightweight, JSON-based format and associated APIs that enable:

1. Declaration of sacred, flexible, and forbidden traits
2. Persistence of creative identity across sessions
3. Compilation into model-specific prompts (SDXL, DALL·E, etc.)
4. User control over what is remembered and how it’s used

Core Concepts

• Sacred Traits: Non-negotiable visual/conceptual anchors (e.g., “emerald left eye”)
• Contextual Modifiers: Scene-specific details (e.g., “rainy night”)
• Negative Constraints: Explicit exclusions (e.g., “no helmet”)
• Provenance: Source of each trait (user-specified, consistent across generations, etc.)

Example CIF Snippet

{
  "cif_version": "0.1",
  "subject": "Kaela",
  "sacred": [
    { "trait": "jade goggles", "weight": 1.4, "locked": true },
    { "trait": "twin braids", "weight": 1.3 }
  ],
  "contextual": [
    { "trait": "desert temple", "weight": 1.0 }
  ],
  "negative": [
    { "trait": "hat", "weight": 0.2 }
  ],
  "user_consent": "explicit"
}