Shot//swarm
the hybrid vfx/ai production hub
Shot//swarm keeps every prompt, plate, camera, version and approval attached to the shot it belongs to, so AI output becomes production work you can direct, review and hand off.
Generative AI and traditional VFX grew up as two separate worlds. One is fast but hard to control. The other is precise but slow and expensive. Studios need both, and today nothing connects them.
Fast, inventive, hard to control
Precise, directable, slow
And in the gap between them, the work falls apart:
Artists jump between AI generators, DCCs, review apps, folders, spreadsheets and chat. Nothing holds the shot together.
Source frames, prompts, model settings and approvals get lost the moment a file is saved.
Raw generations are hard to turn into VFX assets, review packages or an editorial cut.
Most AI tools ignore sequences, USD, ACES, DCC integration and studio versioning.
Shot//swarm fuses the two worlds.
AI output becomes VFX data you can solve, light, comp and deliver. VFX data, like cameras, depth, sets and lighting, steers the AI. One shot, one context, both worlds.
In Shot//swarm the shot is the unit of work. References, takes, spatial data, notes and deliveries stay attached to it from the first look frame to the final EXR.
Shot//swarm works in both directions. It generates media from production context, and it reconstructs production context from media. Choose a step to see what runs there.
Each workspace is a view on the same shot data, so creative, layout, comp and production never work from different versions of the truth.
Studios need to know what was generated, with what, at what cost and with whose consent. Shot//swarm records it per output, by default.
Every output keeps its source frames, prompt, model, provider, seed and settings. Re-render an approved shot with a newer model and compare old and new side by side.
C2PA credentials on delivered files, a likeness consent register per character, and a per-shot report of what was generated and what was shot.
Estimate time and cost per shot and sequence before generating, set GPU cloud spending caps, and bid with real numbers.
See every shot, version, reference and review state in one place.
Control assets, scene context, operators and production artifacts.
An open, standards-based architecture that plugs into real pipelines.
Structure, review and AI workflows without enterprise overhead.
Scout, block and shoot virtual coverage before the set exists.
A whole studio pipeline for one person. Generate, solve, light, comp and deliver on your own GPU, while versions, references and lineage are kept for you. Clients get professional review and delivery, not a folder of renders.
Test Shot//swarm on a real sequence, with our team alongside yours. We agree the success criteria before we start and measure them together.
Plan prices will be announced at launch. Founding Members can lock in their price today.
One subscription. No monthly token bills.
You pay for Shot//swarm and nothing else. Local models run on your own GPU at no extra cost. The only extra costs are optional and up to you: an external paid model, or cloud GPUs when you want more power than your own machine. Both are billed through your own accounts, at the provider’s price, with no markup from us.
Local-first AI/VFX workflow for individual professionals.
For professionals shipping to clients and pipelines.
Shared projects, roles and compute for production teams.
Private deployment for studios with governance needs.
Reserve a Founding Pro license today and pay nothing until launch. You get beta builds before release, a direct line to the development team, and your founding price locked for as long as your subscription stays active.
Prices in USD, excluding VAT. No payment is taken when you reserve: you are charged at launch, and you can cancel at any time before then. Plans and features may change before general release.
Only the Shot//swarm subscription. There are no Shot//swarm credits or tokens to buy. Models that run on your own GPU cost nothing extra.
Only if you choose to. If you use an external paid model (for example a cloud video or image model) or rent cloud GPUs, you pay that provider directly through your own account and API keys, at their price, with no markup from us.
Local models and real-time rendering run best on a recent NVIDIA RTX GPU. Without one you can still work with cloud models and cloud GPUs through your own accounts.
Shot//swarm is model-agnostic. It works with local open models and with cloud providers, including LTX, WAN and FLUX, image and video models through ComfyUI, and local or remote LLMs and VLMs. New models are added as operators, so your workflows don't depend on a single vendor.
Shot//swarm is local-first: shows, assets and outputs stay on your machines or your studio server. Nothing is sent to a cloud service unless you choose a cloud model or GPU for a task.
Each model has its own license. Shot//swarm records which model, provider and settings produced every output, so you can check usage rights shot by shot and share a clear report with clients.
You reserve up to 3 seats at $29/month or $290/year with no payment today. We email you before launch and you are charged only then. You can cancel at any time before launch, and your price stays locked for as long as your subscription stays active.
Yes. Studio / Team plans include a production server, and Enterprise adds fully on-premise deployment with advanced permissions and audit.
25+ years as lighting and FX technical director on feature films, game cinematics and commercials. Gnomon Workshop instructor and long-time Autodesk beta tester.
Co-founder of Digital Video, maker of Toonz, the 2D animation software used on Studio Ghibli films and Futurama, with 4,000+ licenses in 66 countries. Three decades building and selling professional media software.
Models change every month. Production context compounds. We are building the vendor-neutral layer that makes AI usable inside professional pipelines, and we are talking with investors and strategic partners who want to build it with us.
Shot//swarm is in active development, with new builds every few days. This is what landed in the private alpha recently.
Render the shot camera with NVIDIA RTX, real-time or fully path traced, with Beauty, HDR, Albedo, Normals and Depth passes. Renders become shot versions you can A/B against the plate, next to a stable interactive Hydra viewport.
A MaterialX Standard Surface editor with presets and a show-wide library. Materials edited in the Shot Workspace now travel with props to Virtual Cinema and Unreal Engine.
A rebuilt Shot Workspace where plates, cameras, models, lights and AI layers form one USD shot scene. Import USD, glTF, FBX, OBJ, point clouds and HDRIs, then send the whole scene to Virtual Cinema in one step.
Bokeh traced through a real lens, surface by surface: spherical aberration, coma, true cat's-eye, per-wavelength color fringing and lens vignetting. Tilt-shift with Scheimpflug focus, diffraction, GPU tracing, a lens library and Zemax import.
Character controls change the lens itself, not the image. Traced internal ghosts with real coatings, matte box and hood, custom aperture shapes, and dust and scratches that show inside the bokeh.
A traced anamorphic adapter with cylindrical surfaces, anamorphic flare and lobed bokeh, and per-pixel scatter rendering with no tile seams.
Track CineCameras, VCam and Live Link subjects from Unreal at up to 120 Hz, see a live comp, and save the current pose as a shot preset.
A location brief built from script and panels, real-world search on OpenStreetMap and Mapillary, and sun position for your shoot dates.
Automatic checks on generated clips before they reach review.
Lens distortion with ST maps, animated lenses and a real-time lens view in the viewer.
Real-ESRGAN and SeedVR2 upscaling, plus RIFE frame interpolation on the GPU.
A linear compositor with plate integration, layered PSD export and local AI generation up to 8K.