3 months ago
And the winner is...
The judges have weighed in, and we're excited to share the news! Go to the Backblaze Generative Media Hackathon: Build with Genblaze on B2 gallery to congratulate the winners and check out all of the awesome submissions.
Even though the competition is over, you don't have to stop coding. Update your portfolio to inform your followers about new projects and get feedback from fellow hackers.
Questions?If you have any questions about the hackathon, please post on the discussion forum.
3 months ago
Final stretch, Backblaze builders - you've got this!
Dear Hackers,
Whatever stage you're at right now, get something demoable. A rough end-to-end run beats a polished half-pipeline every time, so if you're deep in a hole on one provider or one step, it's a great moment to swap in a fallback model, simplify the ask, or just ship what's working. Judges want to see what you built and why it matters, not a flawless system.
If you're still deciding what to build, or looking for a faster path from pipeline to deployable app, check out our experimental AI SaaS starter kit: auth, billing, and a Genblaze-wired backend already…
3 months ago
Final call for submissions
There are only 3 days left to complete your submission for Backblaze Generative Media Hackathon: Build with Genblaze on B2.
3 months ago
genblaze v0.7.0
As you continue building for the hackathon, we have released a GenBlaze SDK update focused on reliability, clearer errors, and opt-in functionality without changing existing pipeline behavior.
Power your generative media pipelines with Backblaze B2, including 10 GB of free S3-compatible cloud storage and free egress up to 3x your average monthly storage.
Providers
- Google: Added GeminiImageProvider, a native Gemini image-generation provider (google-gemini-image), alongside Imagen. Chat and vision calls now correctly translate ImageURLContent into Gemini’s supported wire format. Known but access-gated Imagen models now return OK_PROVISIONAL rather than a misleading OK.
- Runway: Image-required models now return a clear error…
4 months ago
genblaze v0.6.0
genblaze v0.6.0
To help you keep building with confidence throughout the hackathon, genblaze v0.6.0 focuses on SDK compatibility, cross-platform reliability, and verifiable outputs, so your pipelines keep working as provider SDKs move and your assets land where you expect.
Connectors
- ElevenLabs: ported to the elevenlabs 2.x SDK. Timestamped TTS now reads the parsed AudioWithTimestampsResponse correctly, and model errors fall back to your configured alternates instead of failing the step.
- LMNT: ported to the lmnt 2.x SDK (synchronous client with word-level timestamps). Requires lmnt>=2.6.
- GMICloud Seedance: first/last-frame (FLF2V) generation now routes both frames to their native first_frame /…
4 months ago
genblaze v0.5.0
genblaze v0.5.0
To help you build with confidence throughout the hackathon, genblaze v0.5.0 delivers stronger connectors, safer execution, and more reliable pipelines from prototype to production.
Connectors
- Replicate: now supports community models (e.g. sczhou/codeformer, tencentarc/gfpgan) via automatic version resolution.
- OpenAI Sora: updated for the openai SDK 2.x — image→video chaining and downloads use the current input_reference / download_content API.
- Google Veo: full support for Vertex AI mode (project/location auth) alongside the Gemini Developer API.
- GMICloud: now ships type information (py.typed) for downstream type-checkers and validates video duration inputs.
Reliability & correctness
- Streaming is now fully…
4 months ago
Genblaze Multi Provider Starter App
We built genblaze-gen-media-multi-provider-sample, a Genblaze SDK sample app that turns one prompt into a narrated, scored, captioned MP4.
https://github.com/backblaze-labs/genblaze-gen-media-multi-provider-sample
The app demonstrates a full generative media pipeline across storyboard planning, image generation, video generation, text-to-speech, music, and final MP4 composition. Each modality can use a different provider, selected per run from the UI. Supported providers include OpenAI, Replicate, Google, NVIDIA, Decart, GMICloud, Runway, Luma, ElevenLabs, LMNT, and Hume.
Genblaze removes the need to hand-roll provider glue for auth, retries, polling, error handling, asset storage, and lineage tracking. Developers can orchestrate complex media workflows through simple .step() calls while Backblaze B2…
4 months ago
genblaze v0.4.0
New providers: Hume Octave (text-to-speech) and AssemblyAI (speech-to-text — the first connector that turns audio into a hash-verified transcript with word-level timings).
Reliability & cost
- Retries now resume an in-flight job instead of re-submitting, so a transient poll/fetch error no longer triggers a second billed generation.
- Fan-in steps fed by a failed upstream now fail loudly (FAILED/INVALID_INPUT) instead of silently succeeding on empty inputs.
Security & provenance
- Hardened SSRF protection — DNS pinning plus per-hop redirect re-validation on all outbound fetches — so user-supplied media URLs are safe to handle.
- Manifest.verify() now rejects output assets without a sha256, and DALL-E…