One idea in text — a finished clip with sound
Write your idea — Vini takes it through plot, frames and video, then stitches a finished mp4: 9:16 for Reels and Shorts, 16:9 or 1:1. In the cloud, no install, no local GPU. You see the cost before you run.
The path: idea → video in 7 steps
One line of text goes in. A finished video comes out — 9:16, 16:9 or 1:1: scenes from 3 to 30 seconds each (you pick the max), up to 12 scenes. By default Claude writes the text, image models generate the frames, video comes from one of eleven models you choose between, and ffmpeg does the final stitch inside — so you get a single mp4, not a pile of clips. Every step is real: hit Run and an actual generation runs, not a mock.
- 1Idea (text)one line — what the video is about
- 2PlotClaude breaks the idea into acts
- 3Scriptwriterscenes, lines, who is on screen
- 4StoryboardSTART · MID · PEAK — before you pay for the render
- 5Imagethe first frame of every scene
- 6Videothe frame comes alive, sound arrives with it
- 7Assemblyffmpeg stitches one mp4 in your frames’ format
The work happens on desktop — the studio is built for a big screen.
What people make with it
Not a feature list — jobs. Each one is covered by something already working in the studio; where there is a full write-up, the link sits next to it.
A series, not a one-off
Your hero lives in the library and travels between projects: tomorrow’s episode starts with a finished character, not from scratch.
Read more →A conversation on screen
Up to three heroes at once plus an off-screen narrator — and it costs no more than a single character.
Read more →One video, many languages
Dubbing into 16 languages that keeps the voice timbre, with lip sync to the new speech. Some models speak the lines themselves.
A hero from your photo
Upload a photo and get a character who holds the same look across every scene, without describing them in words.
Read more →Finishing on top of the result
Subtitles, titles, overlays and a timeline you edit by hand. Rebuilding after an edit is a render without AI — so, $0.
Read more →Fix a single scene
Third shot came out wrong — you pay for the third shot. Everything you did not touch comes from cache for free.
Read more →Pick the model per scene
Eleven video models in one node: run the scene through several and compare — each price is visible before you launch.
Read more →See the storyboard before spending
First the storyboard sheet and the shot frames, and only then the money for video generation.
Read more →Every step is visible. Every step can be redone, one at a time
This isn't a black box of «prompt → random result». Plot, script, storyboard, every frame and every scene are right in front of you, in nodes with previews. Didn't like the third scene? Regenerate just that one — the rest stay untouched and aren't paid for again. The cache remembers everything unchanged, and stale badges highlight what went out of date after your edits. Plus a history of it all: projects, runs, graph revisions, autosave and a live generation log — not a single result is ever lost.
The result already exists — Run takes it from cache for $0. Want it redone? The button is right there, and it says plainly that this one costs money.
Library: everything you generated at hand, searchable by prompt.
«Generated it twenty times before it worked» — sound familiar?
Prompt → Run → not it. Another Run → almost. Another Run → all over again — and you pay for every run in full, even for the parts that already worked. One result, twenty paid generations.
Here the principle is different: every step is its own node. You see what it produced. You see what re-running it costs. And only it re-runs — everything unchanged is pulled from the cache and costs nothing.
A typical generator
Canvas with cache
The price is up front, not in a bill afterward
Generation spends real money, and we don't hide that — quite the opposite. The projected cost of a run is calculated and shown on the toolbar before you press Run. Real model prices right in the interface: a frame from $0.035 (Seedream 5.0 Lite) to $0.17 (GPT Image 2), Nano Banana 2 — $0.08. Change one prompt and you pay for one step, not the whole pipeline: unchanged nodes come from the cache for free.
Every number on this page and in the interface is what the models cost us. There is no markup on anything. The pricing policy has not been decided yet, and this is the floor: it will not get cheaper. Whoever joins the closed beta now, joins at this VIP price.
A 30-second video from scratch — on the default models
On the cheapest tier (Seedream 5.0 Lite + Seedance 2.0 mini) the same video is about $1.33. That is the cost of generating from scratch, at what the models cost us; redos are counted separately and pull everything you did not change from cache.
Editing one scene — what actually gets charged
The estimate sits right next to Run — before you press it. A screenshot, not a mockup; the $11.14 here is a whole project, not one scene.
Switch the model — the price recalculates
This is exactly how it works on the canvas too: the model changes in the node, the price — before you press Run.
Your character keeps the same face across scenes
The biggest pain of AI video is a character who looks different in every scene. Here the Element and Soul nodes keep things consistent: define the character once and it holds the same look across every scene through multi-reference — up to 14 references with Nano Banana 2. Add your own photos with the Upload node by a simple drag&drop.
The element library turns one-off videos into a series: today's hero carries on tomorrow's episode in a new project — an episode, not a rebuild from scratch. And so you don't have to be a prompt engineer, there's a prompt library: your own plus built-in system ones, including a ready, fine-tuned Scriptwriter.
One source — every scene: check your hero's face in each frame for yourself.
You do not have to assemble it all by hand
The canvas is your ceiling of control, not a required curriculum. Vini lives right next to it: he sees your whole project, explains any node in plain words, and proposes graph edits that you accept with a click. Money is only ever spent on your explicit press — he never launches anything himself.
Vini — your project copilot
He knows your graph and your run history. Ask «why did this scene come out wrong» or «what should I do next» and the answer is about your project, not in general.
Decoration and Montage
Subtitles, titles, overlays and a timeline — edited by hand on top of the AI plan. Rebuilding after an edit is a render without AI, which means $0.
Voice
Pick a voice or clone your own, record the voiceover, dub it into another language, mix it with music. Separate nodes in the same graph.
The Decoration studio: an overlay on the frame, tracks on the timeline and the «render without AI, $0» badge.
The best models, your choice, in one pipeline
The model switches right in the node — the pipeline doesn't change. A better model ships — you just pick it from the list in the same project. Prices are the same as in the interface: not a «from» price, but the cost of that specific generation.
Text and script
Images
Video · per second
Closed beta — seeking partners and testers
Vini Studio isn't a public service yet: it's a working pipeline where every Run is a real generation for real money. Right now I'm opening access selectively — looking for creators and partners ready to try the product and give honest feedback.
Curious to try it or collaborate — get in touch. Sign in with Telegram or Google, projects are isolated per account. If this is your first time, the account is created on your first sign-in.
Sign in ↑✈️ Or drop by the Telegram channel — updates and contact