Start with the world you built.
Planned intake for Godot scenes, GLB/glTF assets, textures and animations. Developers choose what to include and keep original files under their control.
Planned · Godot integration
HashMo InteractiveSoftware · Games · AI Systems
HASHMO SOFTWARE NEURAL GRAPHICS
EARLY DEVELOPMENT · RESEARCH PROTOTYPE
Your game.
Your art.
Your upscaler.
What if your game's graphics could be reconstructed from fewer rendered pixels without losing the look you created? HashMo Neural Graphics Trainer (HNGT) is our research project to help indie developers eventually train game-specific neural upscalers using their own assets and rendered scenes.
The trainer is not available yet. We have a working application foundation, but no completed HNGT super-resolution model or Godot runtime.
Godot firstArt-direction fidelityLocal-first goalCreator-owned models goal
NEURAL GRAPHICS / WORKFLOW CONCEPTPROPOSED
01 · LOW-RES RENDER02 · REFERENCE DETAIL
LOW-RES INPUT
REFERENCE TARGET
01ImportGodot scenes & assets
02CaptureAligned image pairs
03TrainCompact 2× model
04ValidateQuality & frame time
THE PLANNED WORKFLOW
An upscaler that learns your game.
Our goal is to help developers build a small reconstruction model for a specific game's visual identity. The model would learn from aligned low-resolution renders and high-quality reference images, rather than automatically restyling the world toward photorealism.
Pair fewer pixels with full detail.
Capture the same scene and camera pose at low and high quality, preserving alignment, lighting and metadata for meaningful reconstruction training.
Planned · Paired render captureTeach a compact 2× model.
Train a reconstruction network against the developer's own scenes, with explicit checks for geometry, colors, edges and unintended visual changes.
Planned · Neural trainingMeasure what actually improves.
Compare held-out frames against bicubic and Godot's native upscalers, then evaluate motion artifacts, GPU time, memory use and real game performance.
Planned · Quality labMADE FOR INDIE DEVELOPMENT
The look is yours.
The choice should be too.
HNGT is being designed for independent developers who want to improve rendering efficiency while preserving stylized worlds: chibi adventures, hand-painted environments, crisp sci-fi geometry, and everything in between.
Our commercial direction is a buy-to-own local trainer paired with a Godot integration developers can distribute with their own games under reviewed terms. No customer licensing or runtime redistribution terms are finalized yet.
DEVELOPMENT RECEIPTS
Built foundation. Open research.
We separate verified foundation work from future goals so developers know exactly what exists today.
Independent HNGT codebase
Adapted from HashMo Model Factory with HNGT application branding, a separate repository, existing dataset infrastructure and safety gates that disable inherited Prism stylizer training/export paths.
Source snapshot and product UI verified; this is not neural upscaling yet.My Pet Demon
Our Godot 3D chibi/dark-fantasy game is the planned first capture and training experiment. The game itself remains separate and its assets are read-only research inputs pending rights review.
Godot capture and training results not yet available.2× model and Godot runtime
The paired Godot capture adapter, trained super-resolution model, real-time GPU inference, and developer-ready Godot plugin remain development milestones.
No supported download, benchmark claims or release date.QUESTIONS WORTH ANSWERING
Research with clear boundaries.
Does HNGT already make my Godot game run faster?
No. There is not yet a trained HNGT upscaler or integrated Godot runtime. We will only make performance claims after real game frame-time and visual-quality acceptance.
Can I train directly on 3D or 2D assets?
The proposed trainer starts from assets, but a screen-space upscaler needs matched low-resolution and high-quality rendered images. The future capture tools will generate those pairs from scenes, views, animations and lighting conditions.
Will it change my game's art style?
Preserving the original art direction is the central design goal. Learned reconstruction can still make mistakes, so we plan held-out scene comparisons and motion tests rather than promising perfect reconstruction.
Will players need to buy HNGT to play games made with it?
Our intended licensing model is for developers to buy the creator tool and redistribute a compatible runtime with their games under approved terms. No public price or redistribution license has been finalized.
STAY INFORMED · COMING LATER
Interested in building your own upscaler?
HNGT is in early research. Its planned neural upscaling and Godot capture features are not yet ready for downloads or orders.
Development preview only. Further technical milestones will be published as they are verified.