Quantum Brush: NISQ-Enabled Digital Painting

• Quantum Brush is an interactive system that transforms user strokes into shallow quantum circuits for digital painting and image manipulation.
• It employs local processing and variational quantum algorithms to enable nuanced color evolution and geometric control.
• Beyond digital art, Quantum Brush extends to quantum rendering and polymer-brush thermodynamics, showcasing diverse applications.

Searching arXiv for papers on "Quantum Brush" and related topics to ground the article in current literature. Quantum Brush denotes, in the contemporary arXiv literature, a family of interactive computational systems that map user actions such as strokes, region selections, or local image edits into executable quantum circuits whose measured outputs are rendered back into color or structure. In its primary usage, the term refers to an open-source digital painting environment with NISQ-compatible brushes, later extended with variational quantum algorithms for geometric control and chemistry-inspired color evolution (Ferreira et al., 1 Sep 2025, Lu et al., 30 Dec 2025). In a distinct and non-artistic usage, “Quantum Brush” also appears in polymer-brush thermodynamics, where a discrete energy eigenstructure and a steepest-entropy-ascent equation of motion are used to model non-equilibrium solvent–brush kinetics rather than interactive graphics (McDonald et al., 2023).

1. Scope, origin, and conceptual basis

The digital-painting lineage of Quantum Brush originates with Ferreira et al. (2025) as an open-source tool in which strokes are translated into shallow quantum circuits designed for simulator and real-NISQ execution. Its stated purpose is to make quantum dynamics and constraints visible through painting by exposing superposition, entanglement, interference, measurement-induced randomness, decoherence, and the no-cloning theorem as operative image transformations rather than metaphors (Ferreira et al., 1 Sep 2025).

The underlying design principle is local, stroke-wise quantum processing. A user selects a brush and parameters, draws a stroke or lasso-selects regions, and the application constructs a circuit whose parameters are derived from local color statistics and brush controls. The circuit is then executed on a simulator or on hardware such as IQM’s Sirius, after which measured observables or reconstructed single-qubit states are mapped back into pixel updates. This architecture is explicitly NISQ-aware: algorithms are shallow, local, and formulated so that hardware noise can remain aesthetically legible rather than being fully suppressed (Ferreira et al., 1 Sep 2025).

A second, later strand generalizes this framework by replacing fixed brush dynamics with variational quantum algorithms. In that formulation, one brush learns control fields that steer a source quantum state toward a target state, while another replays the circuit sequence of a VQE procedure for $\mathrm{H}_2$ so that colors evolve according to a physically motivated energy landscape. This places Quantum Brush within a broader class of hybrid interactive systems in which a classical UI orchestrates quantum-state preparation, variational updates, and reconstruction back to image space (Lu et al., 30 Dec 2025).

2. Core digital-painting brushes

The initial Quantum Brush system defines four brushes, each organized around a different quantum effect, and the variational extension adds two more (Ferreira et al., 1 Sep 2025, Lu et al., 30 Dec 2025).

Brush	Quantum mechanism	Visual function
Aquarela	Entanglement-mediated blending	Watercolor-like interpolation
Heisenbrush	Trotterized Heisenberg dynamics	Magnetization-driven color evolution
Smudge	Amplitude damping/pumping cascade	Erasure and uneven decoherence
Collage	Universal asymmetric quantum cloning	Imperfect regional copying
Steerable	Variational quantum geometric control	Source-to-target region steering
Chemical	VQE-inspired circuit family	Chemistry-shaped color evolution

Aquarela encodes the mean hue and luminosity of each stroke segment into qubits and couples them sequentially to an ancilla brush qubit. Conditional $R_z$ and $R_y$ rotations steer segment colors toward the brush color with user-controlled strength $\gamma$, while a controlled $R_y(\pi/3)$ progressively weakens later interactions. The result is a shallow, ancilla-mediated entanglement pattern whose single-qubit tomography produces updated hue–luminosity values, with saturation preserved classically (Ferreira et al., 1 Sep 2025).

Heisenbrush treats the stroke as a quantum spin chain. For $N$ qubits, capped at $10$, it evolves the initial product state $\ket{\psi(0)}=\prod_{i=1}^{N}\ket{(\phi,\theta)}$ under the Hamiltonian

$$\hat{H} = \frac{1}{2}\sum_{n=1}^{N}\big( -\hat{X}_n \hat{X}_{n+1} - \hat{Y}_n \hat{Y}_{n+1} - \hat{Z}_n \hat{Z}_{n+1} + \hat{X}_n + \hat{Z}_n \big),$$

approximated by first-order Trotterization with $\Delta t=0.1$ and $R_z$0. Color is then modulated by the magnetization observable

$R_z$1

through the update rule

$R_z$2

with $R_z$3 (Ferreira et al., 1 Sep 2025).

Smudge implements an amplitude-damping or amplitude-pumping channel with a shared ancilla that is not reset between segments. Operationally, controlled $R_z$4, CNOT, and $R_z$5 gates transfer information into the ancilla so that later qubits interact with a nontrivial ancilla state. This creates an explicitly nonuniform “quantum cascade,” rather than a uniform local blur, and the measured outcome may display both darkening consistent with dephasing and bright emergent accents generated by the cascade structure (Ferreira et al., 1 Sep 2025).

Collage is region-based rather than stroke-based. It encodes a selected image patch through the singular values $R_z$6 of its RGB matrix and prepares a copy qubit with

$R_z$7

A three-qubit universal asymmetric quantum cloning circuit then produces reduced density matrices

$R_z$8

subject to

$R_z$9

with symmetric optimum $R_y$0. The artistic consequence is structurally recognizable but necessarily imperfect copying (Ferreira et al., 1 Sep 2025).

3. Encoding, tomography, and image reconstruction

The painting-oriented Quantum Brush framework uses a consistent color-to-qubit map based on HSL coordinates. Hue $R_y$1 is converted to azimuth $R_y$2, luminosity $R_y$3 to polar angle $R_y$4, and saturation $R_y$5 remains classical to avoid mixed-state preparation on NISQ hardware. A segment or brush color is therefore represented by

$R_y$6

which establishes a direct Bloch-sphere semantics for interactive color manipulation (Ferreira et al., 1 Sep 2025).

After circuit execution, single-qubit tomography reconstructs the updated color state from Pauli expectations:

$R_y$7

The decoded $R_y$8 are then mapped back to hue and luminosity, while saturation is preserved. This encoding–evolution–tomography loop is the basic mechanism behind Aquarela and Smudge, whereas Heisenbrush substitutes magnetization-driven updates and Collage reconstructs image patches through modified singular values and the original singular vectors (Ferreira et al., 1 Sep 2025).

The variational extension broadens the representation repertoire. Steerable uses patch-wise RGBA feature extraction via SVD, building an $R_y$9-qubit state from $\gamma$0 and powers of $\gamma$1 for $\gamma$2 to $\gamma$3 qubits. Chemical instead aggregates hue–lightness angles along a stroke and encodes them with single-qubit $\gamma$4 and $\gamma$5 rotations. These choices preserve compactness and interactivity while keeping the quantum stage compatible with small-$\gamma$6 devices and simulators (Lu et al., 30 Dec 2025).

4. Variational quantum brushes

Variational Quantum Brushes place VQAs at the center of the brush behavior. Their common backbone is a parameterized state

$\gamma$7

whose parameters are optimized against artistic objectives evaluated through quantum measurements. Two distinct objectives are emphasized: fidelity-driven steering for image merging, and energy minimization for chemistry-inspired color evolution (Lu et al., 30 Dec 2025).

Steerable formulates image blending as a quantum control problem. Given source and target regions encoded as states $\gamma$8 and $\gamma$9, one learns control amplitudes $R_y(\pi/3)$0 in the Hamiltonian

$R_y(\pi/3)$1

so that the propagated state approaches the target while minimizing control effort. The paper states the objective as a fidelity-to-target term plus a quadratic control-energy regularizer, written generically as

$R_y(\pi/3)$2

The drift is Heisenberg-like,

$R_y(\pi/3)$3

with cyclically assigned Pauli controls, and the time-ordered evolution is approximated with a second-order Trotter splitting at timestep parameter $R_y(\pi/3)$4, default $R_y(\pi/3)$5. Reported examples include Renoir steered toward a red parrot, Warhol diptychs, and Miró motifs, with a noted tendency toward gray near $R_y(\pi/3)$6 in some settings (Lu et al., 30 Dec 2025).

Chemical reinterprets VQE trajectories as color dynamics. A molecular Hamiltonian for $R_y(\pi/3)$7,

$R_y(\pi/3)$8

is paired with a DUCC ansatz and precomputed offline across bond distances $R_y(\pi/3)$9. At paint time, the application loads the nearest stored circuit family $N$0, applies successive circuits to qubit registers encoding stroke-local hue–lightness angles, and reconstructs updated colors from measured $N$1. The paper reports that smaller bond distances in the $N$2 tests produced more color variability, and compares the resulting dynamics against Aquarela, Heisenbrush, and Smudge (Lu et al., 30 Dec 2025).

These variational brushes do not simply add optimization to the original tool; they change the semantics of the brush. Steerable exposes quantum geometric control as interpolation and extrapolation in image space, while Chemical exposes the temporal structure of a ground-state search as an evolving pigment. The implementation is available open-source and is stated to be fully compatible with the original Quantum Brush application (Lu et al., 30 Dec 2025).

5. Recognition and rendering extensions

The phrase “Quantum Brush” is also used, or is a plausible extension, in work that is not primarily about digital painting but provides adjacent architectures for interactive sketch recognition or rendering. One such direction is the hybrid classical–quantum architecture for vectorized sketch classification. There, sketches are represented as sequential strokes

$N$3

and compressed through Bézier-based, recurrent, and fully connected processing to a $N$4-dimensional feature vector $N$5 matched to $N$6 qubits. A single-layer HEA with angle encoding and $N$7 readout then supports three-class QuickDraw recognition, with QuantumDraw reporting average validation accuracy $N$8 and the separable variant $N$9 on the cellphone/camera/calculator subset (Cordero et al., 2024).

This suggests that a recognition-oriented Quantum Brush can be constructed by coupling live stroke ingestion to a small variational circuit, so that the tool returns recognition or guidance in real time rather than directly modifying paint. The same source explicitly argues that vectorization lowers dimensionality, reduces qubit requirements, and avoids the intractability of pixel-array encoding on NISQ-like resources. It also introduces an important caveat: the no-entanglement variant remained competitive, while freezing the quantum parameters reduced validation accuracy to $10$0, which indicates that joint hybrid training mattered more than entanglement per se in that small task (Cordero et al., 2024).

A second adjacent direction is photorealistic rendering through Quantum Radiance Fields. That work does not define a painting tool, but it gives a plausible rendering substrate for a stroke- or ray-based Quantum Brush by combining quantum encoding circuits, parameterized quantum circuits, a quantum activation function, and quantum volume rendering. Its reported figures include, for Synthetic-NeRF, PSNR $10$1, SSIM $10$2, LPIPS $10$3, and $10$4 FPS, with dense angle encoding and CRz-based circuit families favored in ablations (Yang et al., 2022). A plausible implication is that Quantum Brush can refer not only to color-evolution brushes on an existing canvas, but also to quantum-accelerated generation or revealing of scene content along user-controlled rays or stroke paths.

6. Ambiguity, limitations, and non-artistic usage

A common misconception is that Quantum Brush denotes a single standardized algorithm. In fact, the literature supports at least three technically distinct uses. The primary usage is the digital-painting tool with four fixed brushes and later variational extensions (Ferreira et al., 1 Sep 2025, Lu et al., 30 Dec 2025). A secondary, inferential usage treats compact hybrid QML or quantum-rendering systems as candidates for interactive brush-like interfaces (Cordero et al., 2024, Yang et al., 2022). A third usage is non-artistic: the SEAQT “Quantum Brush” description of polymer brushes in solvents (McDonald et al., 2023).

In the polymer-brush setting, the “quantum” qualifier refers to a quantum-thermodynamic formalism over a discrete energy eigenstructure, not to quantum image synthesis or painting. The system is defined by energy levels $10$5, degeneracies $10$6, and occupation probabilities $10$7, with entropy and energy

$10$8

and a SEAQT equation of motion that generates a unique non-equilibrium kinetic path toward equilibrium. In this formulation, structural observables such as radius of gyration, tortuosity, polymer density, solvent density, brush height, and width are expectation values over the evolving probability distribution, and the predicted density profiles for polystyrene in cyclohexane agree qualitatively with neutron-reflectometry-based measurements (McDonald et al., 2023).

The digital-painting literature also contains several explicit limitations. The original tool emphasizes local, shallow circuits and does not report shot counts, device fidelities, or error-mitigation protocols; indeed, it deliberately uses unmitigated noise as part of the aesthetic, which makes reproducibility backend-dependent and leaves Collage especially noise-sensitive because of its exponential remapping from measured expectations to singular values (Ferreira et al., 1 Sep 2025). The variational extension is similarly constrained by small qubit counts, lossy encodings such as RGBA-SVD and HL-angle maps, and the need for offline precomputation in the Chemical brush (Lu et al., 30 Dec 2025).

Taken together, these strands define Quantum Brush less as a single software package than as a research program for interactive quantum-mediated image manipulation. In its strictest sense, it names a NISQ-compatible digital painting tool and its variational descendants; in broader usage, it names a class of interfaces in which local visual operations are delegated to small quantum circuits; and in a separate thermodynamic literature, it names a non-equilibrium model of polymer brushes whose “quantum” content is formal rather than visual (Ferreira et al., 1 Sep 2025, Lu et al., 30 Dec 2025, McDonald et al., 2023).