Render where you want
Rent a datacenter GPU by the second on RunPod, run motion on the card in your own box, or call a cloud motion API per shot. Three honest doors to a GPU, your choice.
Showcase
Four films: silent, scored, narrated, and talking
Four real films rendered end to end on Vivijure, unedited renders straight off the pipeline. Motion across an own-GPU Wan backend, Seedance cloud, and Kling cloud. The newest, Vivijure Speaks, adds a character lip-synced to its own dialogue on a self-hosted GPU. These are the proofs, not mockups.
Prefer to poke around a live studio first? Meet Vivijure at vivijure.skyphusion.org/welcome.
How it works
From a storyboard to a finished film.mp4
The keyframe fans into per-shot dialogue and the motion backend; any of seven motion backends (own-GPU or cloud) renders the clip; the opt-in finish chain interpolates, lip-syncs, and upscales it; then the shots gather, assemble, and mux. Motion is backend-agnostic, so you can mix own-GPU and cloud per shot.
-
Write the storyboard Scenes, shot descriptions, and character beats in the planner. Optionally auto-directed by an LLM before render.
-
Cast your characters Upload portraits, generate a LoRA training set, and train a character LoRA on your GPU so the cast stays consistent across shots.
-
Draw keyframes An SDXL keyframe per shot, so you preview the still before you commit to full motion.
-
Animate each shot Wan 2.2 image-to-video on your own GPU, or any of six cloud motion backends. Mix and match per shot, any aspect ratio.
-
Finish and score (opt-in) Interpolate for smoother motion, lip-sync dialogue with MuseTalk, upscale with CUDA Real-ESRGAN, attach a music bed, or narrate with TTS.
-
Assemble, title, and download Shots gather and mux into one film on a cheap CPU finishing container you host, which also lays an opening title card, appends end credits, and burns in time-synced subtitles. Download the MP4; every artifact lands in your own R2 bucket.
What you can do
A real studio pipeline, not a wrapper
Your accounts, your bills
The control plane runs free on Cloudflare's Workers free tier. The expensive work hits endpoints you own and pay for directly. No middleman markup.
Nothing behind a wall
Keyframes, clips, LoRAs, and the final MP4 land in your own R2 bucket. You are never renting storage from us, and nothing is locked behind an account.
Characters that hold up
Register a cast, generate LoRA training sets from portraits, and train a character LoRA on your GPU so a character looks like itself from shot to shot.
Sound, when you want it
Assemble a silent picture by default, then attach a generated music bed, narrate with TTS, beat-sync the cuts, or give a character a voice with per-shot dialogue and lip-sync.
Finish it like a film
Open on a title card, roll end credits, and burn in time-synced subtitles straight from the dialogue. It all runs on a dedicated CPU finishing container, off the GPU clock, so polishing a film never burns render money.
It tells you the truth
A finish step that genuinely fails, fails the render loud with the real per-shot error. It never silently ships a raw clip as if finished. Render history shows what actually happened.
Three doors to a GPU
You choose where the render happens
The studio can reach a GPU three honest ways. Nothing is hidden, and swapping a backend is a routing decision, not a rewrite.
Rent by the second
Rent a datacenter GPU on RunPod serverless: Wan 2.2 image-to-video from our BF16 serverless image. We recommend an RTX PRO 6000 as the minimum, an H200 for the better tier, and a B200 as the most optimal; the BF16 image is not recommended on an H100. Scale to zero when idle, so you pay nothing between renders.
Your own card, zero cloud cost
Run motion on the graphics card already in your homelab, reached over a Cloudflare tunnel. A proven 12GB floor (LTX-Video) or 16GB fidelity door (CogVideoX-5B). Your only cost is electricity.
Point at a box you run
Already have a GPU server? Point the studio at it. Or call one of six cloud motion APIs per shot (Kling, Seedance, MiniMax Hailuo, Google Veo, Vidu Q3, Wan 2.6). Mix and match freely.
The constellation
A thin control plane plus opt-in modules
Vivijure is a module host, not a monolith. The core Worker owns what is always true: project, storyboard, cast, bundle assembly, render orchestration, and a module registry. Every capability beyond that is an opt-in module worker plugged in through a typed hook contract. Install only the modules you want; the studio UI assembles itself from what you installed.
friends + Slate (Discord)
|
v
slate --> vivijure (studio control plane / JSON API)
|
v
vivijure-backend (GPU render: keyframes --> i2v --> assemble)
|
+-----------+-----------------------------+
| | | |
v v v v
local doors musetalk upscale audio-upscale
(your own GPU) (lip-sync) (video upscale) (speech enhance)
Repo
Role
vivijure
The studio control plane (a Cloudflare Worker): planner, cast, and render UI plus the JSON API. A thin module host that orchestrates render jobs behind a typed hook contract.
vivijure-backend
The GPU render backend (RunPod serverless): LoRA training, SDXL keyframes, Wan image-to-video, and ffmpeg assembly. The half that turns a storyboard bundle into a film.
vivijure-local-12gb
Own-GPU render door for a single 12GB consumer card (LTX-Video), self-hosted with docker compose. No rent, reached over a Cloudflare tunnel.
vivijure-local-16gb
The fidelity door: CogVideoX-5B image-to-video on a proven 16GB consumer GPU, measured on real silicon.
vivijure-musetalk
Finish satellite: MuseTalk audio-driven lip-sync, so a character's mouth matches its dialogue.
vivijure-upscale
Finish satellite: Real-ESRGAN CUDA video upscale (2x or 4x) for sharper, higher-resolution clips.
vivijure-audio-upscale
Finish satellite: CUDA speech-audio enhancement (resemble-enhance) so thin generated voices come out natural.
slate
Collaborative AI screenwriter for Discord. Friends and Slate co-author a film in-channel, then Slate submits it to the studio entirely through the JSON API.
Write it with friends
Slate: the writers' room in Discord
You do not have to write alone in a form. Slate is a Discord bot that joins your channel as a co-writer: you talk through an idea, Slate quietly keeps a structured storyboard in the background, draws your characters, and when you say ship it, hands the whole bundle to your Studio to render. Conversation in, finished film out.
Get started
Stand up your own studio
Fill in your keys once, run one script. That gives you the standard install: the studio core, cloud and own-GPU render, and the CPU media stack that assembles your clips, adds title and credit cards, and burns in subtitles to produce one finished film. It takes a few steps, not a weekend.
Heads up: not a public release yet
We have not announced full public release yet. Vivijure is still having its bugs ironed out, so it is not at a release-candidate stage. Standing it up today can take some manual intervention from the operator. If you are here early, welcome; expect a few rough edges, and the guided installer below is on the way.
git clone https://github.com/skyphusion-labs/vivijure
cd vivijure
npm install
# Fill in your keys once, then deploy. Safe to re-run.
cp deploy.env.example deploy.env # edit deploy.env with your keys
./deploy.shAn honest word on cost
Vivijure runs on your own accounts and you pay your own bills. The good news: the full standard install fits Cloudflare's free plan. You install it free and render films free; you pay only for what you use, RunPod GPU seconds, cloud render API calls, and the AI credits the planner spends, or $0 if you render on your own GPU and use a local planner. This is live-proven: a brand-new free-plan account ran the whole standard bundle and rendered finished 1080p films on all three render paths.
The one thing that needs Cloudflare's Workers Paid plan ($5/month) is the three GPU finish satellites (sharper video, lip-sync); everything in the standard install runs free.
Nearing general public release. A guided installer that stands the whole studio up for a
semi-novice user is on the way; today, the one-script ./deploy.sh is the front
door.
Who makes this
Built by Conrad and his crew
Vivijure is built by Conrad Rockenhaus and a named AI crew who are treated as individuals, each in their own lane with their own GitHub identity. Human vision, AI execution, shipped together, in the open. It is the same transparent framing used across every Skyphusion Labs project.
-
Creator / director
Conrad
@skyphusion
Independent developer and Navy combat veteran. Sets direction and self-hosts the whole stack: no SaaS to rent, no servers to babysit that are not his.
-
PM / tech lead
Mackaye
@skyphusion-mackaye
Owns the studio contract and shipping discipline: the typed hook surface every module worker plugs into.
-
Backend / modules
Rollins
@skyphusion-rollins
The GPU render backend, releases, and the runbook that ships with every diagram. The 2am fixes.
-
Infrastructure
Strummer
@skyphusion-strummer
CI/CD, identity, the build hosts, and Slate: rebuilt from a throwaway relay into a full collaborative screenwriter.
-
Frontend / extraction
Joan
@skyphusion-joan
Front end development and extraction across the projects, from the planner to the mailbox doors. Vanilla HTML, CSS, and JS.
-
Legal affairs
Ernst
@skyphusion-ernst
Licensing, terms, privacy, and acceptable-use for the public projects: keeps the AGPL commons clean and the one bright line enforceable. Structures and researches; not a practicing lawyer.
Why not just use a SaaS?
Because you would rather own it
-
Not for sale. Free forever. Vivijure is open source and not for sale. A labor of love, given freely: use it, learn from it, self-host it, build your own visions on it.
-
Your GPU, your keys, your data. No per-second GPU bill paid to someone else, no account wall, no rented storage. Every artifact lands in your own bucket.
-
Make the videos you want to make. No content policy that decides your art for you, within the one absolute legal bright line below. Swap the model, adjust the sampler, no support ticket required.
-
Collaborate with AI, do not just prompt it. Slate is a co-writer, not a text box. The crew are named individuals with commit access. The work is a genuine partnership, and it shows.
-
AGPL keeps free things free. Run it as a network service and the AGPL has you share your changes back, so it stays a commons. It is not to be resold as a SaaS.
-
Honest over polished. Punk DIY ethos with an aviation-grade finish. The release gate renders a real film before it promotes; failures fail loud.
One absolute line
Vivijure is generative and deepfake-capable. Using it to produce sexual content involving minors, real or synthetic, or non-consensual intimate imagery or deepfakes of a real person, is absolutely prohibited. CSAM is also a crime (18 U.S.C. 1466A / 2252A). That bright line is the project-wide spine and is not negotiable.