TLDR
We've shipped significant upgrades to Fiddl.art's AI video and integration capabilities. Our new MCP server lets ChatGPT Apps and other clients use our tools with the same schemas and billing as our HTTP API. Sora v2 and v2 Pro now support exact start-image preparation with smart fallbacks. Text-to-Speech streams over SSE with lower latency. We've also added OAuth 2.1, TikTok server-side events, and edge-rendered SEO pages.
MCP Server: Standard Tools, Zero Surprises
Our new Model Context Protocol server enables MCP-compatible clients like ChatGPT Apps to call the same tools our web and mobile apps use. The transport uses SSE plus JSON-RPC with streaming support where beneficial.
Key features include tooling parity with browse_creations, create_image, create_video, and media purchasing functions. The same input schemas, billing rules, and ownership checks apply as our HTTP API. Clients can fetch our OpenAPI specification for self-configuration, and authentication uses existing JWT tokens for consistency across services.
OAuth 2.1 integration includes discovery, dynamic client registration, PKCE authorization, and rotating hashed refresh tokens with minimal consent UI. Tokens sign with our existing keys to maintain stack compatibility.
Sora v2 and v2 Pro: Reliable Start-Images and Faster Pipelines
We've added first-class support for Sora v2 and Sora v2 Pro with intelligent start-image handling. The system automatically handles exact sizing requirements: 720×1280 and 1280×720 for Sora v2, 1024×1792 and 1792×1024 for Sora v2 Pro.
When aspect ratios match, images undergo simple resizing. For mismatched aspects, the system generates a conservative outpaint billed as standard image generation. JPEG format ensures decoder stability, and deterministic caching avoids redundant processing.
Provider integration now includes streaming uploads for large outputs and dedicated status polling. Stall detection automatically errors jobs stuck at high completion percentages for over 15 minutes. We've also implemented soft-delete timestamps and daily purging of originals older than 30 days to optimize storage.
Streaming TTS Over SSE
Our Text-to-Speech service now streams via Server-Sent Events, significantly reducing latency. Clients receive metadata first, followed by audio chunks, progress events, and completion signals. Typed events simplify client implementation with clear event categorization and termination handling.
Authentication and Security Enhancements
Authentication now uses a single JWT token for both HTTP and MCP access. Public reads work without tokens, while writes require ownership verification. OAuth storage hashes client data, auth codes, and refresh tokens with PKCE enforcement where applicable.
OpenAPI generation occurs at startup to maintain consistency between MCP discovery and HTTP surfaces. We've included a CLI inspector example to streamline local testing workflows.
Analytics and Attribution Improvements
We've enhanced our analytics infrastructure with TikTok server-side event integration. The system captures attribution tokens client-side, attaches them to purchases and CompleteRegistration events, and sends server-side events for improved match rates.
Pixel implementation now uses new Meta Pixel IDs with safer initialization and centralized event management. Telegram Mini Apps feature a dedicated runtime that avoids third-party scripts when embedded, properly handles viewport sizing, and integrates Stars purchase flows with native invoices.
SEO and Documentation Updates
New edge-rendered pages for Events and Promo Code Claim include JSON-LD FAQ and HowTo markup with controlled caching. Sitemaps now incorporate new pages and recent public media for improved image and video discovery.
Documentation features multi-language examples and router-registration guidance. Our OpenAPI specification reflects new endpoints and enumeration updates for better developer experience.
Creator and Team Features
Custom model training maintains end-to-end support with Sora start-image preparation compatibility. The model randomizer and multi-select workflows now provide accurate cost estimates for both public and private runs.
Discount code functionality allows buyers to apply codes with upfront validation. Administrators can create, update, list, and delete codes while issuing affiliate payouts with receipts.
Quality of Life Improvements
Comment functionality now includes full CRUD operations, mentions, counts, and notifications with NSFW consent overlays for safe browsing. Training sets feature client-side WebP conversion and thumbnailing with file limits and size guards.
Administrative statistics provide live counters and compact charts for monitoring usage and revenue trends across the platform.
Getting Started
For MCP integration, fetch the OpenAPI resource from our MCP server and register tools with your existing JWT in the Authorization header. Sora implementations simply need to supply a start image—the system automatically handles resizing or cached outpaints to target resolutions.
TTS streaming requires connecting to the SSE endpoint and handling metadata, audio, progress, and completion events sequentially. Discount code validation should occur before checkout to ensure accurate USD and points calculations.
Future Developments
We're tracking upcoming work in our public development boards. While no new cards have been added to Up Next or In Progress since this release, we'll share updates as these lists evolve. Our engineering team continues to enhance Fiddl.art's capabilities for AI art creation and custom model training.
Why These Updates Matter
These enhancements make Fiddl.art easier to integrate, faster to iterate with, and more transparent to operate. The MCP server eliminates glue code, Sora pipelines deliver predictable and cost-aware processing, and streaming TTS reduces latency significantly.
Marketing teams benefit from improved attribution, administrators gain clearer controls, and creators access more reliable tools for AI art generation with custom models. Whether you're building an AI art playground or face forge generator, these updates help you move faster with fewer surprises.
Frequently Asked Questions
What is the MCP server and how does it work with ChatGPT Apps?
Our Model Context Protocol server allows ChatGPT Apps and other MCP-compatible clients to access Fiddl.art's tools using the same schemas and billing as our HTTP API. It uses SSE plus JSON-RPC transport with streaming support where beneficial.
How does Sora v2 handle start images with incorrect aspect ratios?
When aspect ratios don't match Sora's requirements, the system generates a conservative outpaint billed as standard image generation. For matching aspects, it performs simple resizing. Both approaches use deterministic caching to avoid redundant processing.
What advantages does streaming TTS offer over previous implementations?
Streaming via Server-Sent Events significantly reduces latency by allowing playback to begin with the first audio chunk. The typed event system (metadata, audio, progress, done, error) simplifies client implementation and provides clearer termination handling.
How does the TikTok server-side event integration improve attribution?
The system captures attribution tokens client-side, attaches them to relevant events like purchases and registrations, and sends server-side events to TikTok for improved match rates and more accurate conversion tracking.
Can custom AI models work with Sora video generation?
Yes, our custom model training system maintains full compatibility with Sora start-image preparation. Creators can train models using our Forge system and use them seamlessly with Sora v2 and v2 Pro pipelines.
