AI for production and post-production / Operational · under active development

AIArtisan / AiA logo

AIArtisan / AiA

Native AI for the New Era of Craftsmanship.

AiA is the first AI-native Maison built to bridge the gap between pure code and pure art. We don't simply add AI to an existing workflow — we use it to rebuild the VFX pipeline from the ground up, placing the artist back at the center.

Born in Europe and refined for the global stage, AiArtisan is a high-grade intelligence engine that masters the technical friction, so you can focus on the soul of the frame.

Advanced Generative AI. Professional-Grade VFX. Artistically Handcrafted.

AI generationMaskLabColor ScienceProvenance
Media

Footage, plates, references, generated images, clips, image sequences, masks and alpha.

Operations

AI inference and local operators, with explicit input roles, preparation and execution controls.

Automatic provenance

The Graph records the work as it happens, supporting creative exploration and production review down to available shot costs.

From the artist’s work to the production pipeline

AIArtisan brings generative models, media preparation, masking, colour management and review into one project environment. The starting point is the creative process: building an image, developing alternatives, working on a shot and deciding which version to take forward. AiA is designed to handle repetitive technical preparation and maintain continuity between stages, leaving the artist in control of the work.

AiA sits between source material and compositing, editorial, colour and review tools. A plate can be prepared, associated with a mask and processed for cleanup; a generated image can become a reference or the first frame of a video operation.

The operational areas share tenant, project and shot context. Derivatives remain distinct from originals and reusable in subsequent operations. The operator approves preparation that changes the material and explicitly starts AI processing.

  1. 01 / IN

    Material and intent

    Footage, plates, concepts, references and shot direction.

  2. 02 / AiA

    Traceable processing

    Preparation → generation / cleanup / matte → comparison → versions.

  3. 03 / OUT

    Continue in post

    Images, clips and sequences for compositing, editing, color and review.

Current exchanges with post-production tools use exported media and formats.
Work context: assets, projects, shots, runs and versions
Asset
An item in the project: an original, generated image, clip, sequence, mask, extracted frame or conversion. It can be inspected and reused.
Tenant, project and shot
The tenant identifies the working context; the project holds material and processing; the shot connects assets belonging to a particular shot.
Run
An identifiable execution with inputs, a model or operator, prompts, settings, status and outputs. It explains how a result was produced.
Reference, ROI and mask
A reference guides content; an ROI marks a region of interest; a mask identifies where to act or which subject to isolate, according to the operation.
Version and Hero selection
A derivative remains separate from its original. Labels, notes and Hero selection help identify and choose the asset to take forward.
Provenance
The network of relationships between sources, runs and results. The Graph shows both where an asset came from and where it was used next.

Operational paths

Generation, cleanup, mattes and colour preparation

Four paths through generation and media processing. The sections work on the same assets, retaining sources, derivatives and relationships. Open a workflow to inspect its steps.

Build a scene
  1. 01Reference
  2. 02Run Lab
  3. 03Gallery / A–B
  4. 04Extract
  5. 05Run Lab

Generate an image, explore alternatives, animate a frame and extract a new still to use as a first or last frame in the next scene. Prompts, references and steps remain connected.

Clean up a shot
  1. 01Footage
  2. 02Edit / SAM
  3. 03Propagate / Edges
  4. 04ProPainter
  5. 05Gallery / Graph

Start with footage, select a region, prepare its mask and use temporal cleanup. Compare the result with the source and retain the chosen version in the project.

Prepare a matte
  1. 01Footage + mask
  2. 02Cutie · optional
  3. 03Trimap
  4. 04ViTMatte
  5. 0516-bit alpha

A mask can be propagated through time or refined directly when one is already available for each frame. The trimap guides uncertain edges; the resulting alpha returns to Gallery linked to its footage.

Prepare an input
  1. 01Log / linear plate
  2. 02Color Science
  3. 03Input check
  4. 04Prepared version
  5. 05Run Lab

Interpret camera material, choose the encoding for the task, compare previews and scopes, then save an identifiable copy. The model receives the prepared input while the original stays available.

Workflows are assembled by the operator. Local checks and previews precede explicit AI execution.

Workspace sections and shared tools

Home

Project identity and access to the working areas; tenant/project context and service status in the shared shell.

Ingest

Import multiple independent files or image sequences. Assign shots, folders, tags, names and notes before ingest; hashing, metadata and previews prepare the assets for the project.

Gallery

Sources, generated media, masks, exports, data and archive. Shot, folder/collection, date, model and tag filters; bulk organization of selected assets, input selection and direct reuse.

Search

Search across assets and metadata: names, prompts, tags, models and context. Semantic and hybrid search when the embedding engine is configured.

Run Lab

Model/provider selection, prompts, references, input roles, masks and ROI. Model-specific controls, validation, input preparation and run submission.

Runs / Jobs

Job state and progress, results, errors and available actions for each operation. Run inspection, explicit replay and recovery of result publication.

MaskLab

Four connected areas: Edit with Bézier and brushes, Edges for cleanup, Propagate for Cutie tracking and Alpha for ViTMatte. Frames, ranges and sources stay associated.

Color Science

Encoding, gamut/gamma, ACES/OCIO, LUTs, preview monitoring and formats. Scopes, comparison and identifiable saved color versions.

Graph

Global, ancestry and downstream-use views for an asset. Shot/run grouping, filters, search and links to Gallery and Inspector.

Costs

Project and tenant totals, API/model breakdowns, measured usage, estimates and reconciled costs. Account credit and usage reports remain separate from attributed spending.

Settings

Encrypted tenant/project credentials, configuration status, input transport and public URL when needed; preview-maintenance tools.

Inspector, player, Extract and session tools
  • Inspector: previews, technical data, prompts, notes/tags, organization, sources, masks, results and contextual actions.
  • Unified player: stills, videos and image sequences; zoom/pan, frame stepping, loop and side-by-side, split, overlay or A/B hold comparison.
  • Mixed comparisons: video/sequence, still/video and still/sequence within the same viewing system.
  • Extract: a frame or In/Out range, usage note, identifiable badge and reuse in Run Lab.
  • Hero selection to highlight a version, replay with edited settings and editable masks with Bézier, layers, brush and eraser on the selected frame.
  • The download center, contextual guides, jobs bar and persistent page context support everyday work.

Production graph

Graph: automatic tracking, review and production oversight

The Graph builds itself automatically as work is carried out in AiA, recording sources, inputs, runs, results and relationships between versions. Artists can explore and iterate without first drawing a pipeline or tagging every step. Tags, notes and naming conventions enrich the context; technical links are derived from the operations performed.

Global, Lineage and Used By views reveal origins and downstream uses: from plate to prepared copy, from masks to cleanup or matte, and on to colour version and export. Shot/run grouping, filters and search make complex paths inspectable. Nodes link to Gallery and Inspector for asset details and available actions; the same matte can feed multiple operations while retaining its footage association.

Content identity and operation identity remain distinct: hashes recognize identical data; the run retains model, provider, inputs, prompts, parameters and results. Reusing an asset does not erase its operation history. Hero selection identifies the chosen version while alternatives and derivatives remain inspectable.

Source range, reference frame and temporal order also belong to these relationships. An extracted frame, clip or matte sequence retains its position relative to the source. Linking shots, runs and usage records lets production trace operations and available costs for each shot, distinguishing measured, estimated and reconciled amounts.

Pipelines that emerge from the work can be examined and revised separately from the creative session, to compare alternatives, evaluate steps and identify reusable work. This is AiA’s artist-first principle: technical and production oversight draws on the work history without requiring artists to organize every experiment for reporting. Future automation of established paths is a development direction; today the Graph makes their structure and provenance inspectable.

Generation and transformation

RunLab: inputs, models and execution

RunLab exposes controls specific to the selected integration: prompts, references, input roles, ROIs and masks. Text-based generation, semantic editing, inpainting, first/last frames and motion control are distinct workflows, available according to model capabilities. The catalog identifies the function and service for each integration.

Before execution, input checks cover asset type, format, resolution, duration and colour information. When conversion or sequence extraction is needed, AiA proposes a preparation plan for operator approval. Runs / Jobs collects status, progress, results, errors and available actions: inspection, explicit replay and recovery of result publication.

MaskLab

MaskLab: selection, editing, propagation and alpha

MaskLab has four connected areas: Edit, Edges, Propagate and Alpha. SAMURAI uses a region of interest for segmentation and tracking; SAM 3 can start from a text description. The resulting mask can be corrected with brushes and Bézier curves or built manually. Layers, additive/subtractive operations, undo and overlays support frame-level intervention.

Contour extraction produces editable points and handles while retaining the raster base for soft edges, transparency and fine detail. Fitting reports correspondence and deviation from the mask: curves describe the edited frame; Cutie propagation works on the raster mask, not animated splines.

Edges includes threshold, inversion, noise and hole cleanup, expand/contract, feather and levels. Alpha builds a trimap with foreground, background and uncertain edges for ViTMatte, which returns continuous 16-bit PNG alpha. Editing and edge refinement run locally; tracking and matting are started as AI runs.

A practical example

Operational case: bottle removal

The source is Full HD variable-frame-rate footage. This case uses reference frame 62 and range 0–100: 101 masks aligned with source frames. The bottle mask is corrected in the editor and propagated with Cutie in both directions; footage and masks then pass to ProPainter.

Preparation retains correspondence with the original frames. When an operation requires CFR, conversion is proposed explicitly with a chosen frame rate. Range, reference and temporal order identify the processed portion; backward-tracking results return to chronological order. The same mask sequence can continue to Edges or Alpha.

The details show source, propagated mask and High Quality result on the same frame, without subsequent retouching. ProPainter masters are 8-bit RGB; editable mask saves and ViTMatte output use 16-bit alpha. Reconstruction is being refined: the table reflection outside the bottle mask requires a separate operation.

Color Science

Color Science: interpretation, transformation and output

Color Science separates four stages: source interpretation, working space and transform, output encoding/format, and monitor display. Source space, primaries, transfer, YUV matrix, range and alpha are read from available metadata or declared by the operator; detected information remains distinct from assumptions.

ACES 1.3 and 2.0 configurations, scene-linear spaces, display/view, looks and LUTs are handled through OCIO. Configurations with dependencies and 1D/3D LUTs can be imported and recalled through project presets. Supported image and sequence paths include half/float EXR, 8/16-bit PNG, 16-bit TIFF and 10/12/16-bit DPX; video outputs include ProRes 4444, FFV1 and MP4 according to the selected path.

Example: a Log plate is interpreted and converted to an SDR copy suitable for the model. The high-precision original remains available; the AI result returns as a new asset with its own encoding. Converting an 8-bit result to a 16-bit container does not recreate lost precision or dynamic range. Configuration, resources and parameters accompany the saved version.

Waveform, RGB parade, histogram and vectorscope declare their measurement domain and scale. During playback they follow preview samples; when paused, managed measurement can analyze every pixel of the decoded frame. The CIE xy/u′v′ diagram compares primaries and white points of selected spaces: it describes theoretical gamuts, not pixel distribution or HDR colour volume.

Image controls

Adjustments, managed preview and multi-monitor workspace

Exposure, relative temperature, tint, contrast and pivot, saturation, RGB balance and highlight shoulder belong to the saved transform. The panel starts neutral and provides reset and bypass. Preview, GPU/CPU processing and export share the OCIO processor; the float path uses a compatible GPU with CPU fallback in Auto mode.

The current managed preview is SDR/sRGB, including when the master contains HDR values. Where exposed by Windows/NVIDIA, the Display panel reads monitor, mode, refresh rate, RGB/YUV format, range, signal bit depth and ICC profile. These readings describe the workstation and do not change calibration.

AiA can run as an installed web app. Preview, comparison and scopes open in independent windows, including on different monitors, while following the session selection and frame. Position and size are retained with authorized monitor access; docking returns a panel to the workspace. The same system serves the player, mask editor, trimap and ROI.

Costs and services

Costs and Settings: usage, accounts and services

Costs distinguishes spend attributed to a shot, project or tenant, API/model breakdowns, measured usage, estimates and reconciled amounts. Rates carry a source and validity period; currencies remain separate. Available account credit is distinct from the cost attributed to an operation.

Fal and BytePlus support financial-information updates with the configured access; Modal exposes workspace usage reports. Observations carry a period and date: reporting can follow execution and does not automatically constitute a balance or final invoice.

Settings stores encrypted credentials in tenant/project context and distinguishes service configuration from billing-information access. A configured key alone does not establish a successful generation test or available credit. The screenshot shows configuration status without displaying saved keys.

Compare and review

Shared player: comparison and review

The player handles images, video and sequences, including comparisons between different asset types. Side-by-side, split, overlay and temporary A/B, frame stepping, loop, zoom and pan support source/result inspection. Compared assets retain their context and temporal references.

Inspector collects technical information, prompts, notes and tags, organization, sources, masks and results. Extract saves a frame or In/Out clip with a note and provenance, making it available as a reference or first/last frame in a subsequent operation.

Media preparation and continuity

Technical operations around AI inference

Asset preparation, constraint checks and result collection belong to the operational environment. Decisions that change the material are declared and approved by the operator.

  1. 01

    Check the input

    Reads available metadata and checks roles, media type, format, resolution, duration and color information against endpoint constraints. Sequences, videos, masks and references have distinct functions.

  2. 02

    Prepare where you work

    When required, proposes format/color conversion, resizing or sequence extraction. The operator approves the plan and follows progress. Prepared copies retain links to originals; an associated sequence can be reused.

  3. 03

    Work on the right time range

    CFR conform provides a constant cadence when VFR material needs it. In/Out bounds operations that support it; backward processing prepares the useful range without manually exporting another clip.

  4. 04

    Collect and realign

    Results are collected and grouped with metadata and source links. Backward-tracking output returns to chronological order while preserving correspondence with original frames and the source range.

  5. 05

    Reuse or prepare delivery

    Save a precise frame or clip from the player, with a note and provenance, for use as a reference or first/last frame. Color Science prepares the encoding and format of the export version; Gallery retains the alternatives.

Cycle between AI selection, manual editing and further processing
  1. AI
  2. Mask
  3. Bézier / brush
  4. Corrected mask
  5. AI
Ten operations without new AI inference
Identity and media
Content hashes, unique references, technical metadata, thumbnails and proxies make files selectable and identifiable across the workspace.
Check before submission
The model determines input roles, counts and types. Checks compare format, geometry, precision and available metadata; proposed conversions are visible.
Traceable local preparation
Format/color conversion, frame extraction and coordinated source/mask preparation create linked copies. The operator approves the plan and receives progress feedback.
Select useful contributions
Intentional frame and In/Out clip extraction from the player, with original frame/range, notes and intended use.
Transfer and collection
Input packaging for the selected service, transfer through configured channels, output collection and Gallery publication with the corresponding relationships.
Sequences and packages
Frame ordering, timing preservation, provider-package inspection, RLE/PNG masks and grouping outputs under the same run.
Preview and inspection
A shared player, A/B comparison, edge/trimap previews, OCIO conversions, scopes and technical readouts use local processing.
Versions and continuity
Each saved transform records its origin. Persistent drafts, selections and activities help work resume after navigation or refresh.
Export and delivery
Original downloads, sequence packages and dedicated exports. A queue shows preparation, transfer and available actions.
Accounting
Usage measurements, time-bounded rate sources, per-run snapshots, currency-specific totals and reconciliation of available charges.

Local operations use workstation resources. Transfers, storage and external services follow project configuration; technical operations do not all require a new model call.

Documented catalog · 2 October 2026

Models and operators by function

35 integrations in the documented edition: 33 AI endpoints and 2 local operators. AiA connects Replicate, Fal, BytePlus ModelArk and Google Gemini alongside models hosted in the dedicated Modal workspace. Each integration has specific input roles and controls.

AiA’s integration architecture is designed to make it straightforward to add new models and providers through their APIs. Ongoing catalog updates are planned in response to production needs and the evolution of models and services.

Images & editing 11
Seedream 5.0 FlashBytePlus ModelArk
Direct generation/editing and layer operations through a distinct model profile from Pro.
Seedream 5.0 ProBytePlus ModelArk
Direct generation/editing, multiple references, layer decomposition and alpha-image editing.
Nano Banana 2 · Gemini 3.1 Flash ImageGoogle Gemini
Direct Google integration for generation, reference-driven editing and prompt-described local changes.
Nano Banana 2 · GenerateFal
Prompt-based image generation with composition, format and output controls.
Nano Banana 2 · EditFal
Semantic editing guided by prompts and multimodal references supported by the endpoint.
FLUX.2 MaxReplicate
Still generation and reference-driven variations, with aspect-ratio and resolution controls.
Seedream 4.5Replicate
Image creation and editing, multiple references, image formats and output variations.
Nano Banana Pro · Gemini 3 Pro ImageGoogle Gemini
Direct Google integration for images and reference-based editing, with 1K, 2K and 4K sizes.
Nano Banana ProReplicate
Image generation and editing with multiple references, composition and prompt-guided variations.
FLUX Fill ProReplicate
Masked inpainting and outpainting: replacement, cleanup and canvas extension.
FLUX Fill DevReplicate
Masked editing and iterative variations, with generation settings and output-count controls.
Video & motion 9
Seedance 2.5 / DreaminaBytePlus ModelArk
Text, first/last frames, image/video/audio references, editing and extension; draft/final workflow where supported.
Seedance 2.0 · Reference to VideoFal
Generation guided by image, video and audio references assigned to their respective roles.
Seedance 2.0 · Image to VideoFal
Animation from a starting image, with an optional final frame to guide the result.
Seedance 2.0 · Text to VideoFal
Clip generation from a text description, with duration and aspect-ratio controls.
Kling V3 Pro · Motion ControlFal
Transfers motion from a reference video to a subject in a source image.
Kling O3 Pro · Reference to VideoFal
Video from references, character elements and keyframes, with named prompt references.
Kling O3 Pro · Video EditFal
Video editing with prompts, reference images and visual elements.
Kling V3 OmniReplicate
Video generation guided by image and video references, with structured prompt references.
Kling V3 VideoReplicate
Video from text or images; first/last frames and multi-shot descriptions with controlled durations.
Masks & alpha 9
CorridorKeyModal
Chroma keying and alpha refinement from footage and a supporting matte.
SAM 3 · VideoReplicate
Video segmentation using the model’s exposed selections and parameters; masks enter the AiA workflow.

Describe the subject to segment in words, for example “bottle” or “person”. The prompt guides segmentation through the video; request per-frame masks in a ZIP or an overlay visualization with chosen color and opacity. A negative prompt can specify objects to exclude.

SAM 3 · ImageReplicate
Image segmentation to create masks for editing and compositing.

Name the subject to isolate in an image using a text prompt. The resulting masks can be inspected, corrected or used for editing; a mask associated with a footage frame can become a propagation reference.

BRIA · Video background removalReplicate
Extracts a subject from video, with background, audio and alpha-capable container controls.
SAMURAIReplicate
Subject segmentation and tracking with ROI, In/Out preparation including backward input, a consistent reference and reusable masks in chronological order.

Indicate the subject with a box on the reference frame: SAMURAI derives its segmentation and follows its motion. The ROI is the initial guide; the resulting masks describe the subject over time.

BRIA · Image background removalReplicate
Removes an image background and returns a transparent PNG cutout.
851 Labs · Background RemoverReplicate
Image subject extraction and background removal.
CutieModal
Mask propagation over a chosen In/Out range, an arbitrary reference inside that range, forward/backward or both directions, corrections and linked sequences.

Start from a mask drawn in MaskLab or produced by a model and linked to the correct frame. Refine it with Bézier or brush tools, choose the reference and In/Out, then propagate forward, backward or in both directions. The result is a mask sequence linked to its source.

ViTMatteModal
Mask refinement using imagery and a trimap; continuous 16-bit PNG alpha.
Temporal cleanup 1
ProPainterModal
Object removal from aligned source footage and masks, quality profiles and temporal controls. Verified Full HD path; 8-bit RGB masters. AiA’s object-removal workflow is in development, with reconstruction quality being refined.
Depth 2
Depth Anything V3 · MetricReplicate
Scene-depth estimation from an image for preparation and depth-based processing.
Depth Any VideoReplicate
Depth estimation for video material, with outputs associated with their source.
Upscale 1
Real-ESRGANReplicate
Image upscaling with the model’s scale-factor and processing options.
Local operators 2
Mask edge refinementLocal
Local cleanup, morphology, expansion/contraction, feathering and matte-level operations.
Color · OCIO / ACESLocal
Float32 color, display/view, LUT and image trims, with precision controls and traceable output versions.

The same model can appear on different services: each entry describes a distinct integration. Semantic editing and masked inpainting use different inputs and paths. Availability, use and commercial terms depend on the service and account.

AIArtisan / AiA — Anatomy of a Shot Anatomy of a Shot: from text to generated shots, through to the Graph. The film retains the previously presented generation workflow; the screenshots above also show subsequent MaskLab and Color Science developments.

Roadmap

Roadmap: interchange, revisions and delivery

  1. Intent and preparation

    ScriptLab / Scripta

    Script, scenes, breakdowns and planning material: the creative and production context carried through to the shot.

  2. Set and virtual production

    PVT

    Pixera Virtual Tracking: tools for camera, lens and scene states, with control and recall of on-set configurations.

  3. Generation and post

    AiA

    Media, models, masks, color transforms and results: operations that can be traced through provenance.

Shot identity · media · versions · decisions · provenance. An ecosystem vision map: cross-product connections are integration directions.

The roadmap extends a working foundation: deliveries with footage, results and mattes on aligned layers, starting with Premiere and Avid; organized packages with media, frame references and information about the operations.

Interchange directions include OTIO for timelines and connections to version-control systems such as Perforce. Within the KRD ecosystem, ScriptLab / Scripta and PVT outline continuity between narrative preparation, set and virtual production, generation and post.

Context export

Copy the public technical brief.

Public technical context covering workflows, capabilities, the model catalog and integration directions. The brief distinguishes working functions from the roadmap.

This page presents selected capabilities and screenshots. Code, internal architecture, configurations and detailed commercial material are available in the context of a private review.

Collaboration

Discuss the project in more depth.

For collaboration, integration or a closer technical review of AIArtisan / AiA, request the relevant material and start a conversation.

Request access