Backblaze Generative AI Media Hackathon
One character. Every medium. Always consistent.
Loomina turns a single character profile into a full library of images, scenes, voiceover and video — keeping the same face, style and identity across all of them. No model training, no LoRA.
Why it stands out
Consistent identity — without LoRA
The vault's own portraits become the reference. A character keeps the same face across every image, multi-character scene and video — no training, no fine-tuning, in seconds.
One pipeline, end to end
Profile → image modes (single, variation set, photoshoot, story) → studio backgrounds & photo art → voiceover → video. The whole creative chain in one place.
Provenance & AI disclosure built in
Every asset carries a verifiable manifest. Choose a visible “AI” watermark or invisible embedded metadata — transparency by design, not an afterthought.
Waste-aware by design
Draft and final quality tiers, live cost estimates before every run, batch generation with a stop button, and a duplicate-request guard. Spend credits on keepers, not misfires.
How it works
- 1Build a characterGive them a reference image (generated or your own photo), personality, a fixed voice, a seed and a purpose — an extended CV.
- 2Generate in any modeOne portrait, a 10-shot variation set, a 100-frame photoshoot, or a full story series — the prompt builder assembles a strong prompt from optional fields.
- 3Bring them to lifeAdd voiceover, compose multi-character scenes, or animate a portrait into a short video — identity held throughout.
Voice:
Assets
No assets yet — generate one in the Images tab.
Images
Generate portraits, variation sets, photoshoots or story panels for one character — or bring several together in one scene. Pick who's in the frame below.
Pick the character(s):
Genblaze → Backblaze B2 · identity, personality & seed from the profile carry into every frame
Prompt builder optional — style, light, angle…
Assets
No assets yet — generate a portrait above.
Your scenes
No scenes yet — pick characters and describe a scene above.
Studio
Generate images without a character — free artistic photo art like Midjourney, or empty background/scene plates. Same prompt builder, no identity lock.
Prompt builder optional — style, light, angle…
Your studio images
Nothing here yet — pick a mode and generate.
Audio
Turn a script into spoken audio — narration, voiceover, dialogue. Pick a voice from the catalog.
Genblaze TTS → Backblaze B2
Pick the characters (2–6) — each speaks in their own fixed voice:
One line per turn, format "Name: line" — spoken in order, each in the named character's fixed voice.
Your audio
Nothing here yet — write a line and generate.
Your dialogues
No dialogues yet — pick characters and write a script above.
Video
Bring a character to life — animate one of their portraits into a short clip (identity held via image-to-video), or generate video straight from a prompt.
Video generation runs on GMI Cloud and takes 1–4 minutes. It keeps running if you switch views.
No video model, no lip-sync risk — each panel's still scene image is shown for exactly as long as its own line takes to speak. Built for more than one character talking at once, since real animated lip-sync only ever works for one face per clip.
Your videos
No videos yet — pick a model and generate.
Idea → Script
Describe an idea in a sentence and get a ready-to-shoot script. A story script drops straight into the Story image mode — one line becomes one panel.
Optionally include your characters by name:
Saved scripts
No scripts yet — describe an idea above.
Canvas
Bring text and shapes onto your generated images — marketing overlays, manga pages, and more.
My templates
Reopen a saved layout and swap in different images.
No templates saved yet — build a layout and save it from the editor.
Edit
Cut your images, scenes and clips into one film — trim, grade, add camera moves, lay music underneath. The result lands in the vault carrying a manifest of everything it is made of.
The model only orders and times what is already in your vault — it generates nothing, and every shot it picks stays editable.
Preview — looks and camera moves are approximated in the browser; the render is the truth.
Click material on the left to put it on the timeline. Drag the clips to reorder them; click the marker between two clips to switch cut ↔ fade.
No track yet. Anything you add plays under the clips' own sound, not instead of it.
My cuts
Every saved cut stays editable — it only references your assets, so reopening one costs nothing.
No cuts yet — build one under “New cut”.