genblaze-openai

OpenAI provider adapters for genblaze — Sora text-to-video, DALL·E / gpt-image text-to-image, and TTS text-to-speech — with SHA-256 provenance manifests on every output.

genblaze-openai wraps OpenAI's generative media APIs (Sora video, DALL·E 3 and gpt-image-1 images, tts-1 / tts-1-hd / gpt-4o-mini-tts audio) as genblaze providers. Compose them into multi-step AI pipelines, persist outputs to Backblaze B2 or any S3-compatible store, and emit a tamper-evident provenance manifest for every run.

Why genblaze-openai

• Three OpenAI modalities, one SDK — video (Sora), image (DALL·E, gpt-image), audio (TTS) — same Pipeline API.
• Built-in provenance — Every Sora render / DALL·E image / TTS clip lands with a SHA-256-verified manifest.
• Swap models without rewrites — Same pipeline works with Runway, Luma, Flux, Veo, ElevenLabs, etc.
• Production-ready — Retries, timeouts, moderation hooks, step caching, streaming events.
• Durable storage — Plug genblaze-s3 in for B2 / AWS S3 / R2 / MinIO persistence.

Providers + models

Each is registered via entry points (openai-sora, openai-dalle, openai-tts).

Install

pip install genblaze-openai

Quickstart — Sora text-to-video

export OPENAI_API_KEY="sk-..."

from genblaze_core import Modality, Pipeline
from genblaze_openai import SoraProvider

run, manifest = (
    Pipeline("sora-demo")
    .step(SoraProvider(), model="sora-2",
          prompt="A cinematic drone shot gliding over a misty mountain valley at sunrise",
          modality=Modality.VIDEO, seconds=4, size="1280x720")
    .run(timeout=300)
)
print(run.steps[0].assets[0].url, manifest.canonical_hash)

Quickstart — DALL·E text-to-image

from genblaze_openai import DalleProvider

run, manifest = (
    Pipeline("dalle-demo")
    .step(DalleProvider(), model="dall-e-3",
          prompt="A watercolor painting of a cozy bookshop on a rainy evening",
          modality=Modality.IMAGE, size="1024x1024", quality="hd")
    .run(timeout=120)
)

Quickstart — OpenAI TTS

from genblaze_openai import OpenAITTSProvider

run, manifest = (
    Pipeline("tts-demo")
    .step(OpenAITTSProvider(output_dir="output/audio"),
          model="tts-1-hd",
          prompt="Welcome to Genblaze — generative media pipelines with provenance.",
          modality=Modality.AUDIO, voice="nova", response_format="mp3")
    .run(timeout=60)
)

Persist to Backblaze B2

from genblaze_core import KeyStrategy, ObjectStorageSink
from genblaze_s3 import S3StorageBackend

storage = ObjectStorageSink(
    S3StorageBackend.for_backblaze("my-bucket"),
    key_strategy=KeyStrategy.HIERARCHICAL,
)
# …then pass sink=storage to .run(…)

See Backblaze B2 — the recommended default sink for genblaze.

Credentials

Documentation

• Main repo: https://github.com/backblaze-labs/genblaze
• Examples: sora_video_pipeline.py · dalle_image_pipeline.py · tts_audio_pipeline.py

Related packages

• genblaze-core — the pipeline SDK
• genblaze-s3 — durable storage on Backblaze B2 and other S3-compatible backends

License

MIT