Genblaze is an open-source Python SDK designed for orchestrating generative AI media pipelines across video, audio, and image providers. It targets engineers and researchers building complex AI media workflows, offering a unified API to abstract provider differences and providing built-in provenance tracking for every output, ensuring traceability and verifiability.

How It Works

Genblaze employs a unified Pipeline and Step API to abstract away the complexities of interacting with diverse generative AI services like OpenAI, Google, Runway, and NVIDIA NIM. This design allows users to swap providers with minimal code changes. A core feature is the automatic generation of a canonical provenance manifest for each pipeline run, which can be embedded directly into media files or stored separately, ensuring auditable lineage and asset integrity. Storage integration with Backblaze B2 and S3-compatible services is a first-class concern.

Quick Start & Requirements

• Primary install: pip install genblaze (installs core and B2/S3 storage). Provider-specific packages are available (e.g., pip install "genblaze[gmicloud]").
• Prerequisites: Python 3.11+. API keys are required for most providers (e.g., GMI_API_KEY, OPENAI_API_KEY, B2_KEY_ID, B2_APP_KEY).
  • Backblaze B2: B2 Application Keys
  • GMICloud: console.gmicloud.ai
  • NVIDIA NIM: build.nvidia.com
  • OpenAI: platform.openai.com
  • Google: aistudio.google.com
  • Runway: dev.runwayml.com
  • Luma: lumalabs.ai/dream-machine/api
  • Decart: platform.decart.ai
  • Replicate: replicate.com/account/api-tokens
  • ElevenLabs: elevenlabs.io/app/settings/api-keys
  • Stability AI: platform.stability.ai
  • LMNT: app.lmnt.com
  • Hume AI: platform.hume.ai
  • AssemblyAI: assemblyai.com/app/api-keys
• Links: PyPI, GitHub CI, Quickstart Example, Architecture.

Highlighted Details

• Provenance by Default: Generates canonical, hash-verified provenance manifests embeddable into media files (.mp4, .png, .jpg, etc.) or storable in cloud buckets.
• Unified Provider API: Supports numerous video, image, audio, and chat providers (e.g., OpenAI, Google, Runway, Luma, ElevenLabs, Stability AI, NVIDIA NIM) via a single Pipeline/Step interface.
• First-Class Storage: Seamless integration with Backblaze B2 and any S3-compatible store for durable asset and manifest persistence.
• Replayable Runs: Manifests capture full provenance, enabling pipeline reconstruction and replayability.
• Fallback Chains & Conformance: Includes mechanisms for retries on model errors and tests to detect upstream provider drift.

Maintenance & Community

The project provides contribution guidelines in CONTRIBUTING.md and repo conventions in AGENTS.md. No specific community channels (like Discord/Slack) or sponsorship details are prominently featured in the README.

Licensing & Compatibility

The project is licensed under the MIT License. This license generally permits commercial use and integration into closed-source applications.

Limitations & Caveats

Genblaze is not intended for simple LLM chat loops (use provider SDKs or LangChain), UI-driven generation apps in JS/TS (use Vercel AI SDK), or scenarios where media generation and provenance are not primary concerns. Outputs that are URL-only, without bytes being transferred or verified, will not pass Manifest.verify().