genblaze-runway 0.3.3

Runway Gen video provider adapter for genblaze

genblaze-runway

Runway Gen-4 / Gen-3 video provider adapter for genblaze — text-to-video AI pipelines with SHA-256 provenance manifests on every render.

genblaze-runway wraps the Runway ML API (Gen-4 Turbo, Gen-3a Turbo) as a genblaze provider. Compose Runway video generations 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-runway

• Runway Gen-4 Turbo, unified API — same Pipeline API as Sora, Veo, Luma, Flux.
• Provenance by default — every render gets a SHA-256-verified manifest with prompt, model, params, timestamps.
• Production-ready — timeouts, retries, progress streaming, step caching, moderation hooks.
• Composable — chain Runway outputs into downstream FFmpeg transforms or AV compositors.
• Durable storage — drop the genblaze-s3 sink in for B2 / AWS S3 / R2 / MinIO persistence.

Models

Model	Notes
gen4_turbo	Latest Runway Gen model — fast, highest quality. Image-to-video only: every request needs an input image (inputs=/external_inputs=/params={'prompt_image': url}).
gen3a_turbo	Previous generation — still supported. Also image-to-video only.
gen4.5, veo3, veo3.1, veo3.1_fast	Support both image-to-video and text-only prompts — with no input image, the provider routes to Runway's text-to-video endpoint automatically.

Install

pip install genblaze-runway

Registers the runway provider via entry points; genblaze-core discovers it automatically.

Quickstart — text-to-video

gen4_turbo/gen3a_turbo are image-to-video only; use a text-capable model (gen4.5, veo3, veo3.1, veo3.1_fast) for text-only prompts:

pip install genblaze-core genblaze-runway
export RUNWAYML_API_SECRET="..."

from genblaze_core import Modality, Pipeline
from genblaze_runway import RunwayProvider

run, manifest = (
    Pipeline("runway-demo")
    .step(RunwayProvider(), model="veo3.1",
          prompt="A timelapse of wildflowers blooming in a meadow, soft morning light, macro detail",
          modality=Modality.VIDEO, duration=8)
    .run(timeout=300)
)
print(run.steps[0].assets[0].url, manifest.canonical_hash)
assert manifest.verify()

Quickstart — Gen-4 Turbo image-to-video

gen4_turbo always requires a source image — pass one via external_inputs:

from genblaze_core import Asset, Modality, Pipeline
from genblaze_runway import RunwayProvider

reference_image = Asset(url="https://example.com/reference.jpg", media_type="image/jpeg")

run, manifest = (
    Pipeline("runway-demo")
    .step(RunwayProvider(), model="gen4_turbo",
          prompt="A timelapse of wildflowers blooming in a meadow, soft morning light, macro detail",
          modality=Modality.VIDEO, duration=10,
          external_inputs=[reference_image])
    .run(timeout=300)
)
print(run.steps[0].assets[0].url, manifest.canonical_hash)
assert manifest.verify()

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,
)
# pass sink=storage to .run(…) — assets + manifest uploaded to B2

Backblaze B2 is the recommended default sink for large AI-generated video.

Credentials

Env var	Where to get it
RUNWAYML_API_SECRET	https://dev.runwayml.com/

Documentation

• Main repo: https://github.com/backblaze-labs/genblaze
• Example: runway_video_pipeline.py

Related packages

• genblaze-core — the pipeline SDK
• genblaze-s3 — durable storage on Backblaze B2 and other S3-compatible backends
• Other video providers: genblaze-openai (Sora) · genblaze-google (Veo) · genblaze-luma (Dream Machine) · genblaze-decart (Lucy)

License

MIT