Hierarchical Project Groups: Subject → Animation → Direction, with persistent assets, state and metadata.
Unum Sunt Sprite Studio
An end-to-end sprite production suite for generating, extracting, cleaning, selecting, aligning, organising and exporting animation assets — with optional managed local AI runtimes.


One production chain, from source material to reusable character assets.
The interface keeps each stage explicit. Work can start from generated media, existing footage or an imported spritesheet, then move through inspection and correction instead of hiding the pipeline behind a single automation step.
WanGP video bridge, image/video references, WAN Animate and the prepared Krea 2 path.
R1 chroma-key Lab, masks, edge decontamination and transparent frame extraction.
Alpha brush, rectangle/lasso selections, propagation and transactional undo per stroke.
Output geometry, pivots, anchors, mirroring, shared scale and per-frame correction.
Smart frame analysis based on MAD / Rec.709, duplicate sensitivity, anomalies and loop quality.
Frames and spritesheets with controlled layout, scale, format and transparency/background.
Productive Presets that capture reusable pipeline settings and apply them to Project Groups.
Calibration Lab for run history, lightweight hardware probes, ratings, A/B comparison and promotion.
Prompt Builder for reusable generation instructions and constrained production prompts.
Import, decompose and normalise existing sheets, then build WAN reference sheets from selected poses.
Separate image-generation provider producing normalised PNG + manifest jobs without coupling the core to the AI runtime.
Guided Standard, Full AI-to-Sprite and Rework routes with checkpoints and explicit step completion.
Character Set / Layer Manager for organising reusable subject assets beyond a single animation.
The suite as it is actually used.
Real development captures from the current production branch. Open any image to inspect the full interface.
Local generation is integrated, but it does not own the application.
Sprite Studio can drive video generation through WanGP/WAN Animate and a separate local image-generation provider. Reference image, motion reference, positive/negative prompts, seed, resolution, frame count, FPS and steps remain visible instead of disappearing into a black-box call.
The core workflow stays usable independently from generation. AI jobs write normalised outputs and manifests, while extraction, cleanup, alignment, selection, spritesheet work and export remain ordinary production stages.
Krea 2 is represented in the runtime-management path, including temporary Hugging Face token handling and explicit licence consent, but the end-to-end Krea 2 setup is still pending after R5c6 and is not presented here as complete.
Inspect, repair and align the frames instead of accepting generation artefacts blindly.
Extraction R1 combines a chroma-key Lab with edge decontamination. Clean-up adds alpha painting, rectangle and polygonal lasso selections, propagation across selected frames and transactional undo for individual brush strokes.
Alignment combines pivots, anchors, mirroring, shared output geometry and per-frame offsets. Smart Selection uses MAD / Rec.709-derived analysis to score movement, near-duplicates, anomalies and sequence quality before the final set is exported.
Existing sheets can be taken apart, normalised and turned back into references.
The Sprite Sheet workspace can import regular grids or irregular atlases, decompose them into frames and re-inject selected material into the production pipeline.
Selected key poses can also be assembled into a WAN Reference Sheet, making reverse-engineering and generation part of the same asset loop instead of separate utilities.
Projects, presets, calibration and guided workflows keep experiments reproducible.
Project Groups organise work as Subject → Animation → Direction. Productive Presets capture reusable settings without overwriting source files or per-frame manual corrections. Guided workflows register which route a group is following and which step comes next.
Calibration Lab stores runs, ratings, useful-frame counts, notes and A/B baselines. A lightweight hardware probe helps compare settings without turning calibration into a hidden benchmark service.
The public build is a real Windows application, not a Python folder with a launcher script.
R5c1 moved the core to a PyInstaller onedir build with LocalAppData configuration/log paths and a frozen-runtime self-check. The build pipeline runs tests, produces a manifest, packages the release and writes a SHA-256 checksum.
R5c4/c4a introduced an Inno Setup installer with Core, Complete and Custom paths. The target machine does not need a pre-installed Python interpreter for the application itself.
Preflight, install, adopt, repair and remove — without taking ownership of someone else’s WanGP install.
The non-destructive preflight checks Windows/CUDA driver capability, writable paths and free disk space, and can detect an existing WanGP installation without changing it. No fixed minimum GPU, VRAM or RAM threshold is imposed by R5c2; effective compatibility is checked against the installed PyTorch/CUDA runtime.
For managed installs, Runtime Manager can bootstrap private Miniconda + Python 3.11.14, PyTorch 2.10/cu130, WanGP requirements and selected models. Downloads are resumable, verified with SHA-256, and downloaded Windows installers are Authenticode-checked before execution.
Existing external WanGP installs can be adopted without moving or renaming their folders. Repair/update and removal only act on managed components. Model cleanup is explicitly constrained to the WanGP checkpoint area, while runtime, models and app data remain independent opt-in choices during uninstall.
Core application first. Local AI only when the machine and the user ask for it.
The public Windows package is distributed through GitHub Releases. Core Sprite Studio is standalone; the heavier local-AI stack is a separate managed layer with preflight, adoption and maintenance tools.
Optional local-AI runtime
Managed Miniconda + Python 3.11.14 + PyTorch 2.10/cu130 + WanGP requirements.
CUDA/driver and installed PyTorch wheel are checked dynamically; R5c2 does not impose a fixed minimum GPU/VRAM/RAM model threshold.
WAN Animate and selected checkpoints can be downloaded into managed model storage with resumable transfer and verification.
May be adopted as external. External installations are never moved, renamed, repaired or removed by managed-runtime maintenance.
Acquisition/licence plumbing exists, but end-to-end Krea 2 setup remains pending after R5c6.
Reproducibility and safety contracts
- Build automation runs tests, writes a manifest, packages ZIP/release assets and generates SHA-256 checksums.
- Runtime downloads can resume and are SHA-256 verified; downloaded Windows installers are Authenticode-checked before execution.
- Hugging Face credentials used for gated acquisition are temporary and are not stored as persistent application secrets.
- Licence acceptance is requested explicitly and is not silently remembered across maintenance operations.
- Uninstall cleanup is opt-in independently for runtime, models and application data.
Automation saves time; the operator still sees the frames.

Distribution, documentation and release status.
Sprite Studio now has a public repository and GitHub Releases. The Windows core is standalone, while the managed local-AI runtime remains an optional layer with preflight, adoption, repair and health-check tooling.
Download
AvailablePublic development packages are distributed through GitHub Releases.
DownloadGitHub
AvailableSource, history and public project documentation are available in the repository.
Open GitHubDocumentation
On this pageThe 14-workspace production chain and validated R5c-series architecture are documented on this page.
Open sectionRequirements
On this pageCore Windows packaging and the optional managed local-AI runtime are documented separately.
Open sectionChangelog
AvailablePublic release history and packaged builds are exposed through GitHub Releases.
OpenSupport
AvailableVoluntary project support routes through the shared GVibeDev Support page; the external PayPal endpoint remains inactive until configured.
Open SupportTutorials
PlannedDedicated public tutorials will be linked when the documentation set is ready.
Planned