vMira Studio gathers our image, video, audio, and 3D models in a single canvas. The point is not to be one more generative tool in a stack of seven. The point is to make the seam between models — the place where teams lose hours every week today — not exist.
The most common question we hear from creative teams is: how do I connect five different tools to produce a single asset? The honest answer, until now, required pasting an image into one tool, animating it in another, voicing it in a third, mixing in a fourth, and exporting for the web in a fifth. Each step lost the prompt that got you there, lost the layer structure of the previous step, and forced a manual round-trip through file formats that were never designed to talk to each other. Studio replaces that loop with a single canvas where every output is also a typed input, every prompt is preserved, and every step is a node you can revisit, branch, or replace.
One canvas, four models
Drag an image output into the input of the video model and it is treated as the start frame, with the original generation prompt available as conditioning. Drag a video output into the input of the audio model and it is treated as scene timing, so a generated voice-over lands on cuts. Drag any output into the 3D model and it becomes a textured plane in a scene you can re-light. Each step keeps the original prompt, the seed, and the parameters — so going back to step one and trying a different camera angle does not destroy steps two through four; they re-render against the new upstream output. The history view is a true graph, not a stack: branches can run in parallel and merge later. Variations of a single shot generate in under a minute on the hosted plan, in roughly four to six on the Pro plan with a dedicated GPU.
Work with what you already have
Studio imports Figma projects with layer fidelity, PSD layer stacks (including smart objects and adjustment layers), .glb and .fbx meshes for 3D, raw DaVinci Resolve project files for video, and any common audio format for the audio model. It exports back to all of the above, plus PNG sequences for compositing pipelines and JSON metadata so your studio can keep track of which prompt, seed, and model version produced which asset. The brief is to complement a creative pipeline, not replace it. If you ship in After Effects, the Studio output is a layered After Effects project. If you ship in Blender, it is a scene with materials. The model output is a deliverable, not a screenshot.
“The goal is not to produce faster. It is to try more versions before committing to one — and to remember which one you chose, and why.”
Typed I/O between models
Every Studio output carries a type. An image is not just pixels; it is pixels-plus-prompt-plus-seed-plus-model-version. When you drag that output into a video model's input slot, the typing is what makes the seam invisible: the video model is conditioned on the original prompt as well as the pixels, so motion stays consistent with the intent of the still. Drag a video output into the audio slot and the audio model receives the timing of cuts in addition to the audio target — a generated voice-over lands on a cut rather than mid-sentence. The typing is opt-in at every step: if you want a clean break, you can pass only the pixels and discard the rest. The default is the smarter behaviour because it is what most teams want.
The history graph
Every step you take in Studio is a node in a graph, not a stack. Branches can run in parallel and merge later. If you generate eight image variants and want to animate three of them in parallel with different camera moves, that is three branches that can converge later when you mix audio across them. Going back to step one and trying a different camera angle does not destroy the downstream nodes — they re-render against the new upstream. The history view is a true graph: zoomable, navigable, exportable as a JSON document that captures every prompt, seed, parameter, and model version. Sharing a Studio session with a collaborator means sharing the graph, not a flat sequence of steps.
Seed propagation and prompt fidelity
A common complaint with multi-step generative pipelines is that small perturbations upstream cause large perturbations downstream. We attack this with explicit seed propagation: when you commit to a step, its seed is locked, and downstream steps inherit it as the basis for their own seed derivation. The result is a session that is reproducible end-to-end. If you re-open the session a week later and re-render, you get the same outputs; if you tweak a prompt at step three, only steps three and downstream change, not the upstream ones. This is more useful than it sounds: it is the difference between a pipeline you can iterate on and a pipeline you have to babysit.
Performance: hosted vs Studio Pro
On the hosted plan, an image generates in two to three seconds, a video in twenty to thirty, an audio clip in five, a 3D mesh in fifteen. Studio Pro adds dedicated GPU acceleration that compresses each of those numbers by two to three times: image in under a second, video in under ten, audio in two, mesh in five. The latency win compounds across a multi-step session — a brand-exploration pass that takes ninety seconds on the hosted plan takes thirty on Pro. For studios with their own GPU farm, the on-premise build is licensed separately and runs the full pipeline locally; latency on the customer's hardware can match Pro or beat it depending on the GPU mix.
Brand templates and locked typography
Studio Pro adds brand templates that lock typography, colour palette, and layout grid for any image or video output. The model's freedom is preserved inside the locked frame — characters, scenes, lighting all remain generative — but the deliverable matches your brand without a manual pass in After Effects. Templates are shared at the studio level, so a junior designer cannot accidentally ship something off-palette, and audit-mode flags any output that touched the locked tokens. The brand-template format is open: customers can author their own and share them between projects, or import from existing Figma libraries.
How Studio composes with the API
Studio is a UI on top of the same models the API exposes. Anything you can do in Studio you can do programmatically — and the export of a Studio session is a JSON document that, run through our API client, reproduces every step. Production teams use Studio for exploration and prototyping, then move the locked-down sessions to the API for repeatable rendering at scale. The two surfaces share auth, rate limits, and billing, so a creative director's exploration in Studio bills against the same budget as the production rendering job — visible in one dashboard.