TavernAI to Backyard AI Converter

The migration from **TavernAI** to **Backyard AI** involves more than a simple format shift; it's a technical optimization for local model orchestration. TavernAI cards utilize auxiliary chunks within a PNG file to store character data, whereas Backyard AI (formerly Faraday) manages a persistent local database of character entities. Our tool bridges this gap by performing a surgical extraction of V2 metadata.

1. V2 Metadata and Chunk Scanning

Standard PNG files are composed of data "chunks." TavernAI cards hide their character "soul" in tEXt or zTXt chunks, often labeled with the chara or ccv3 keys. Our converter performs a binary sweep of the provided image, identifying these chunks and decoding the Base64-encoded JSON strings hidden within. This ensures that even compressed cards, which might fail in simpler tools, are fully parsed and ready for migration.

2. Mapping Logic Normalization

Backyard AI's internal engine is highly optimized for local inference. To ensure your characters perform as intended, we perform a semantic alignment of the TavernAI V2 fields. Elements such as "Post-History Instructions" and "System Prompts"—which are often buried deep in the TavernAI metadata—are meticulously re-mapped to the corresponding fields in the Backyard AI JSON. This prevents the character's behavior from "drifting" and ensures that the model's response constraints are respected.

3. Dialogue and Context Synthesis

Backyard AI thrives on clean, narrative-driven dialogue. We extract the first_mes and mes_example fields from your TavernAI card and normalize them for the local models hosted within the Backyard ecosystem. This includes stripping legacy roleplay artifacts and reformatting dialogue examples into the standard {{char}}: "..." pattern, which provides the model with a clear roadmap for the character's unique speaking style and tone.

4. Privacy and Zero-Server Sovereignty

Absolute privacy is the core of the local AI movement. Mirroring the philosophy of Backyard AI, our converter operates on a "Zero-Server" architecture. Every byte of your character data is processed locally in your browser's RAM using JavaScript. Your creative digital assets are never uploaded, never cached, and never seen by any third party. This ensures that you maintain 100% data sovereignty over your characters as you move them into your private roleplay environment.

5. Optimizing for Model-Specific Performance

Backyard AI supports a wide variety of model formats (GGUF, EXL2). Our tool generates a standardized JSON output that is designed to be universally readable across these loaders. By adhering to the most robust data mapping standards, we ensure that your TavernAI characters are not just "imported," but are truly integrated as high-quality, manageable assets within your Backyard library, facilitating faster inference and deeper narrative immersion.