Desktop ToolDevelopment Preview

Unum Sunt Sprite Studio

A desktop workflow for turning motion references into cleaned, aligned and export-ready sprite assets.

Sprite pipelineWindows DesktopWanGP bridgeFrame extractionCleanup & alignment

One pipeline instead of a folder full of half-finished frames

Unum Sunt Sprite Studio is a Windows desktop workflow built around a recurring production problem: converting a motion reference into sprite assets is not a single operation. Generation, background removal, frame selection, cleanup, alignment and export all create different failure points.

The tool keeps those stages visible and inspectable instead of hiding them behind one “make sprites” button.

The current public status is Development Preview. It is a working development application, but a public repository and packaged release have not been published yet.

What the current workflow covers

Generate with a local WanGP bridge

The Generate workspace can send a reference image and motion reference to a local WanGP installation. Prompt, negative prompt, seed, requested resolution, frame count, FPS and generation steps remain explicit, while job state and output are tracked inside Sprite Studio.

WanGP is an integration, not the whole application. The remaining production stages can be reasoned about and tested independently from generation.

Extract and inspect the mask

Source video frames can be analysed against one or more background colours. The extraction workspace exposes tolerance, edge softening, cleanup and structural mask refinement instead of treating chroma removal as a one-value effect.

Separate views for the original frame, mask, transparent result, detected silhouette and candidate background make failures visible before they propagate.

Clean the result instead of accepting it blindly

Automatic cleanup handles small alpha defects, while the pixel painter and selection tools allow local corrections. Rectangle and polygonal lasso selections can remove unwanted areas without forcing the user to repaint an entire frame.

Align a sequence around useful anchors

Animation frames rarely arrive perfectly registered. Sprite Studio applies a shared output geometry, scale and anchor model, then allows per-frame correction. Onion-skin inspection helps reveal jitter that is easy to miss when looking at frames individually.

Select fewer, better frames

The smart selection stage analyses motion, near-duplicates and anomalies across an interval. The goal is not simply to keep every generated frame, but to propose a compact sequence that better matches the intended animation profile.

Export assets for a game pipeline

Export Studio can produce individual images or sprite sheets with controlled layout, scale, image format and background. The final product is an asset package, not merely a preview inside the tool.

Why the setup is documented publicly

Local AI generation is the part most likely to fail for reasons unrelated to Sprite Studio itself. During development, a seemingly minor Python-version mismatch was enough to make the WanGP bridge report NOT READY.

That is exactly the kind of detail GVibeDev should expose rather than bury. A useful AI-assisted workflow includes dependency management, reproducible environments and diagnostics alongside the visible creative tool.

AI-assisted, but designed around inspection

AI is used where it helps — most visibly in the optional generation stage — while later stages deliberately give control back to the operator. Masks can be inspected, frames can be rejected, anchors can be corrected and outputs can be rebuilt.

The objective is not maximum automation. It is a production chain where automation saves time without making mistakes impossible to see.

In development

Screens from the current build.

Real development captures. Click any frame to inspect the full image.

Generate — reference image, motion reference, prompts, seed, resolution, frames and live job state in one workspace.

Project Groups organize production as Subject → Animation → Direction while preserving local project state.

Extraction R1 starts from the source video and builds a configurable multi-colour background mask.

Structural mask diagnostics expose the detected central silhouette before destructive cleanup.

The transparent preview makes edge contamination and missing pixels visible before later stages.

Clean-up combines automatic alpha repair with pixel painting, rectangle/lasso selection and undoable corrections.

Alignment uses output geometry, shared scale and global/per-frame anchors to reduce animation jitter.

Smart selection analyses a source interval and proposes a smaller frame set for the target animation profile.

Export Studio produces individual frames or sprite sheets with configurable layout, scale, format and background.

Setup / dependencies

Know what the tool needs before you install it.

Sprite Studio can be used as a production workflow, while local AI generation adds a separate WanGP runtime stack. The two layers are documented independently so the Python environment is not hidden behind a generic “install” button.

CategorySprite production workflow

PlatformWindows Desktop

Public packageNot published yet

Core

Sprite Studio

Windows desktop environmentRequired

The current development preview is built and tested as a Windows desktop workflow.

Local generation

WanGP bridge

Miniconda / isolated Python environmentRequired for generation

Recommended for the local WanGP integration so its runtime does not collide with the system Python installation.

Python 3.11.14Required for generation

This is the interpreter version validated for the working local WanGP bridge. Point the bridge to this dedicated interpreter, not to a newer global Python installation.

WanGP StandaloneRequired for generation

Required only for the Generate workspace when using the Local WAN / WanGP Bridge provider. Extraction, cleanup, alignment and export are separate stages.

PyTorch / CUDA stackRequired for generation

The validated local generation machine used PyTorch 2.10.0, torchvision 0.25.0 and torchaudio 2.10.0 from the CUDA 13.0 wheel index.

triton-windowsRequired for generation

Installed after the WanGP requirements in the validated Windows generation environment.

CUDA-capable NVIDIA GPURequired for generation

Part of the validated local WanGP generation setup. The non-generation Sprite Studio stages do not depend on WanGP GPU inference.

Validated development setup

Prepare and launch the validated local WanGP stack

The public Sprite Studio installer is not available yet, so this guide focuses on the dependency chain that previously caused the most trouble: the local WanGP runtime used by the Generate workspace.

Create a dedicated Python 3.11.14 environment
Keep WanGP independent from the global Python installation. The path below mirrors the known-good development layout and can be changed if you keep the same separation.
```
conda create -p C:\AI\envs\WanGP python=3.11.14 -y
conda activate C:\AI\envs\WanGP
python --version
```
Install the validated PyTorch / CUDA packages
This pins the package set used by the working Windows generation environment rather than allowing pip to resolve an arbitrary stack.
```
python -m pip install torch==2.10.0 torchvision==0.25.0 torchaudio==2.10.0 --index-url https://download.pytorch.org/whl/cu130
```
Install WanGP requirements and Triton for Windows
Run these commands from the WanGP Standalone root, the same directory that contains wgp.py and the models folder.
```
cd /d C:\AI\WanGP_Standalone
python -m pip install -r requirements.txt
python -m pip install triton-windows
python -m pip check
```
Verify CUDA before involving Sprite Studio
Confirm that the dedicated environment can actually see the GPU. This isolates WanGP problems from Sprite Studio bridge problems.
```
python -c "import torch; print(torch.__version__); print(torch.cuda.is_available()); print(torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'NO CUDA')"
```
Launch WanGP once on its own
The validated launch used SDPA, profile 4 and FP16. A successful run opens the local WanGP interface, normally on 127.0.0.1:7860.
```
cd /d C:\AI\WanGP_Standalone
C:\AI\envs\WanGP\python.exe wgp.py --attention sdpa --profile 4 --fp16 --open-browser
```
Point Sprite Studio at the dedicated WanGP runtime
In Generate, choose Local WAN / WanGP Bridge. Set Python executable to the dedicated 3.11.14 python.exe, wgp.py to the file inside WanGP_Standalone, WanGP root to that same folder, and select the exported WanGP settings JSON used by your preset. Validate the bridge before starting a generation job.
Keep the working directory contract intact
The bridge must run with the WanGP Standalone root as its working directory. Using another folder can break access to wgp.py, models/_settings.json and relative WanGP resources even when the Python path itself is correct.

This is the configuration validated for the GVibeDev development machine. It documents a known-good environment, not a claim that WanGP can only run with these exact versions. The final Sprite Studio installer/start command will be added when the first public repository and release are published.

Output example

Unum Sunt Sprite Studio

One pipeline instead of a folder full of half-finished frames

What the current workflow covers

Generate with a local WanGP bridge

Extract and inspect the mask

Clean the result instead of accepting it blindly

Align a sequence around useful anchors

Select fewer, better frames

Export assets for a game pipeline

Why the setup is documented publicly

AI-assisted, but designed around inspection

Screens from the current build.

Know what the tool needs before you install it.

Sprite Studio

WanGP bridge

Prepare and launch the validated local WanGP stack

Create a dedicated Python 3.11.14 environment

Install the validated PyTorch / CUDA packages

Install WanGP requirements and Triton for Windows

Verify CUDA before involving Sprite Studio

Launch WanGP once on its own

Point Sprite Studio at the dedicated WanGP runtime

Keep the working directory contract intact

From source motion to game-ready frames.