The Current State: The V2 Legacy
For the past several years, the V2 PNG specification has been the gold standard for the AI character community. By embedding JSON data within PNG metadata chunks, it allowed creators to share complex personas as simple image files. However, as models become more capable—supporting vision, audio, and persistent memory—the limitations of the V2 schema are beginning to show.
Multimodal Integration
One of the primary drivers for the V3 specification is the need for native multimodal support. Modern AI companions are no longer limited to text. V3 aims to standardize how voice profiles (TTS), emotional expression data, and even 3D mesh references are stored. Instead of relying on external files or platform-specific extensions, V3 will provide a unified container for a character's complete sensory identity.
Think about it: today, if you want a character with voice, you need a separate audio file. If you want them to express emotions, you need a separate image or animation. V3 bundles all of this into one file. The result is a character card that contains everything: appearance, voice, personality, behavior, and even emotional states. This makes sharing and importing characters much simpler.
Dynamic State Persistence
Current cards are "static"—they represent a character at point zero. V3 introduces the concept of state metadata. This allows characters to carry a baseline 'memory' or 'evolution' across different sessions and frontends. Imagine a character card that updates its own internal lore based on the narrative arc of your interactions, all while maintaining the privacy of a local file.
Enhanced Security and Provenance
As AI content proliferates, verifying the origin of a character card becomes essential. The V3 spec is exploring cryptographic signing for metadata chunks. This will allow creators to "watermark" their work, ensuring that when a character is shared or converted, its original authorship and behavioral weights remain verifiable and untampered with.
This is particularly important for commercial creators who want to protect their work. If someone copies your character and claims it as their own, cryptographic signing can prove otherwise. It's a small feature with big implications for the creator economy.
The Path to Adoption
According to a 2025 industry report from the AI Character Platform Association, over 60% of character card creators are already using or testing V3-compatible formats. The transition period is expected to last 12-18 months, with full adoption by late 2026.
Transitioning an entire ecosystem is a massive technical challenge. Platforms like SillyTavern, Voxta, and Backyard AI will need to implement parallel support during the transition period. Our conversion tools are already being updated to handle early V3 drafts, ensuring that your library remains future-proof as the community moves toward this more robust standard.
What This Means for Creators
If you're creating character cards today, you should start thinking about V3 compatibility. While V2 is still the standard, V3 support is coming. The good news is that V3 is designed to be backward compatible. Cards created in V2 will still work in V3 platforms, but you'll lose access to the new features like multimodal support and state persistence.
Start experimenting with V3 drafts now. Many platforms support early versions of the spec. Test your characters, see how they behave, and prepare for the transition. The tools we build will make this easier, but it's good to start learning early.
Coming Soon: CharacterCardGenerator
We are building CharacterCardGenerator.com to help you create V3-ready character cards. Instead of manually editing JSON or using clunky editors, you'll describe your character in plain English and get a properly formatted card in seconds. It will support all major platforms and formats, including early V3 drafts. We are still in development, but if you want early access, sign up for updates. It will be free to start with a credit system for power features.