genblaze-s3 0.2.3

S3-compatible storage backend for genblaze (B2, R2, MinIO, AWS)

genblaze-s3

S3-compatible storage backend for genblaze AI media pipelines — durable, content-addressable, dedup-ready. Works with Backblaze B2 (recommended default), Cloudflare R2, MinIO, and AWS S3.

genblaze-s3 plugs into the genblaze ObjectStorageSink to persist AI-generated video, image, and audio — plus their SHA-256 provenance manifests — onto any S3-compatible object store. It handles streaming downloads from provider CDNs, SHA-256 hashing, multipart uploads with retries, pre-signed URLs for private buckets, and Object Lock retention for tamper-evident manifests on Backblaze B2.

Why genblaze-s3

• Durable by default — Assets + manifests land in object storage, never stuck in a provider's expiring CDN URL.
• Backblaze B2 first-class — One-line S3StorageBackend.for_backblaze() helper, Object Lock support for immutable provenance.
• Content-addressable dedup — KeyStrategy.CONTENT_ADDRESSABLE stores each unique asset once by SHA-256.
• Works with any S3 API — AWS S3, Backblaze B2, Cloudflare R2, MinIO, SeaweedFS, Wasabi, Ceph.
• Presigned URLs — private buckets get time-limited URLs; public buckets get permanent public_url_base links.
• Resilient multipart uploads — credential-preserving retries, preflight checks, no partial writes.

Backends

Provider	Helper	Notes
Backblaze B2	S3StorageBackend.for_backblaze("bucket")	Reads B2_KEY_ID / B2_APP_KEY; Object Lock retention supported
AWS S3	S3StorageBackend(bucket="...", region="...")	Standard AWS credential chain
Cloudflare R2	S3StorageBackend(bucket="...", endpoint_url="https://<acct>.r2.cloudflarestorage.com")
MinIO / self-hosted	S3StorageBackend(bucket="...", endpoint_url="https://minio.example.com")

Install

pip install genblaze-s3

Quickstart — Backblaze B2 (recommended)

export B2_KEY_ID="..."
export B2_APP_KEY="..."

from genblaze_core import KeyStrategy, ObjectStorageSink, Pipeline
from genblaze_s3 import S3StorageBackend
from genblaze_replicate import ReplicateProvider

backend = S3StorageBackend.for_backblaze(
    "my-genblaze-bucket",
    # Defaults to "us-west-004". Pass the region your bucket actually lives
    # in (e.g. "us-east-005", "eu-central-003") to skip the redirect hop —
    # the backend auto-corrects on first use, but a right hint saves an RTT.
    region="us-west-004",
    # Optional: pass public_url_base for public buckets (get_url returns permanent URLs)
    public_url_base="https://f004.backblazeb2.com/file/my-genblaze-bucket",
)

sink = ObjectStorageSink(
    backend,
    prefix="genblaze-assets",
    key_strategy=KeyStrategy.CONTENT_ADDRESSABLE,   # dedupe by SHA-256
)

result = (
    Pipeline("b2-demo")
    .step(ReplicateProvider(), model="black-forest-labs/flux-schnell",
          prompt="a photorealistic cat wearing a tiny spacesuit")
    .run(sink=sink, timeout=120)
)

for step in result.run.steps:
    for asset in step.assets:
        print(asset.url, asset.sha256)

backend.close()

Resulting bucket layout with CONTENT_ADDRESSABLE:

genblaze-assets/
├── assets/{sha[:2]}/{sha[2:4]}/{sha}.ext    # one object per unique asset
└── manifests/{run_id}.json                   # one manifest per run

Switch to KeyStrategy.HIERARCHICAL for runs/{date}/{run_id}/… layout (better for run-grouped browsing, worse for dedup).

Quickstart — AWS S3

export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."

from genblaze_s3 import S3StorageBackend

backend = S3StorageBackend(bucket="my-genblaze-bucket", region="us-east-1")
# get_url() returns pre-signed URLs when public_url_base is not set

Quickstart — Cloudflare R2 / MinIO

from genblaze_s3 import S3StorageBackend

# R2
backend = S3StorageBackend(
    bucket="my-bucket",
    endpoint_url="https://<account-id>.r2.cloudflarestorage.com",
    access_key_id="...", secret_access_key="...",
)

# MinIO
backend = S3StorageBackend(
    bucket="my-bucket",
    endpoint_url="https://minio.example.com",
    access_key_id="...", secret_access_key="...",
)

Object Lock for immutable manifests (Backblaze B2)

Genblaze can apply Object Lock retention to uploaded manifests, producing tamper-evident provenance suitable for compliance, legal, and content-authenticity workflows. See the main repo docs for the Object Lock guide.

Documentation

• Main repo: https://github.com/backblaze-labs/genblaze
• Storage feature doc: https://github.com/backblaze-labs/genblaze/blob/main/docs/features/object-storage.md
• Runnable examples: b2_storage_pipeline.py, s3_storage_pipeline.py

Related packages

• genblaze-core — the pipeline SDK
• Provider adapters: genblaze-openai · genblaze-google · genblaze-runway · genblaze-luma · genblaze-replicate

License

MIT