pyimagecuda-studio 0.1.2

Node-based image processing powered by CUDA - Design visually, execute programmatically

PyImageCUDA Studio 0.1.2 Beta

Design image pipelines visually. Automate with Python.

Node-based GPU compositor with real-time preview and headless batch processing.

Core is PyImageCUDA, my custom CUDA library with native kernels for all operations.

Dependencies only for UI (PySide6), I/O (PyVips), and preview (PyOpenGL), no CUDA Toolkit required (just NVIDIA drivers).

https://github.com/user-attachments/assets/6a0ab3da-d961-4587-a67c-7d290a008017

https://github.com/user-attachments/assets/f5c6a81d-5741-40e0-ad55-86a171a8aaa4

---

Key Features

Visual Node Editor

• 40+ GPU-accelerated nodes: Generators, effects, filters, transforms, blend modes (see full list below)
• Real-time preview: CUDA-OpenGL interop for instant visual feedback on every change
• Global variables: Link any parameter across multiple nodes for template reuse
• Flexible controls: Numeric inputs with optional sliders for precise adjustments
• Intuitive workflow: Right-click to add nodes, drag to connect, see results instantly

Headless Automation

• Simple Python API: Load .pics projects and modify parameters programmatically
• Batch processing: Generate thousands of variations in seconds
• Type-safe validation: All parameters validated (colors, sizes, text, ranges, etc.)

from pyimagecuda_studio import LoadProject, run, set_node_parameter

with LoadProject("sign.pics"):
    for i in range(1, 11):
        set_node_parameter("Text", "text", f"Test {i}")
        run(f"output_{i:02d}.png")

Performance

• 10-350x faster: than CPU alternatives for complex operations, see PyImageCUDA benchmarks
• 5-15ms rendering — Typical 1080p pipelines (RTX 3070, varies by GPU)
• Memory efficient: GPU buffers created on-demand and reused intelligently

---

Installation

Requirements

• Python 3.10+
• NVIDIA GPU (GTX 900 series or newer)
• OS: Windows 10/11 or Linux

Install via pip

pip install pyimagecuda-studio

Launch GUI

pyimagecuda-studio
# or shorter alias:
pics

---

Available Nodes

Generators (13 nodes)

• Image — Load from file with resize options
• Text — Rich text rendering (font, size, color, alignment, spacing)
• Dynamic Text — Python-generated text (API calls, calculations, etc.)
• Color — Solid fill
• Gradient — Linear/radial/diagonal gradients
• Stripes — Angled stripe patterns
• Checkerboard — Classic checker pattern
• Grid — Customizable grid lines
• Dots — Polka dot patterns
• Circle — Perfect circle
• Ngon — Regular polygons (3-20 sides)
• Noise — White noise (RGB or monochrome)
• Perlin — Organic noise textures

Effects (4 nodes)

• Rounded Corners — Smooth corner radius
• Drop Shadow — Blur, offset, color control
• Stroke — Inside/outside outline
• Vignette — Edge darkening

Filters (8 nodes)

• Gaussian Blur — Fast separable blur
• Sharpen — Unsharp mask
• Sepia — Vintage tone
• Invert — Color negative
• Threshold — Binary conversion
• Solarize — Partial inversion
• Sobel — Edge detection
• Emboss — 3D relief

Adjust (5 nodes)

• Brightness — Linear adjustment
• Contrast — Midpoint scaling
• Saturation — Color intensity
• Gamma — Non-linear correction
• Opacity — Alpha blending

Transform (6 nodes)

• Resize — Scale to exact dimensions (Nearest/Bilinear/Bicubic/Lanczos)
• Aspect Resize — Maintain ratio while resizing
• Scale — Percentage-based scaling
• Flip — Horizontal/vertical/both
• Rotate — Arbitrary angles with optional expand
• Crop — Extract rectangular region

Merge (2 nodes)

• Blend — Composite two images with Photoshop-style blend modes
  • Modes: Normal, Multiply, Screen, Add, Overlay, Soft Light, Hard Light
  • Positioning: 9 anchor points (top-left, center, bottom-right, etc.) + pixel offset
  • Control: Opacity slider (0.0-1.0)
• Mask — Use image brightness or alpha as transparency mask
  • Modes: Luminance (brightness) or Alpha channel
  • Positioning: Same anchor + offset system as Blend

Logic (2 nodes)

• Random Out — Randomly select between inputs
• Conditional — Python-based routing logic

Utility (1 node)

• Split — Duplicate to 2 outputs

Headless Mode

Design templates in the GUI, automate generation with Python.

Basic Example

from pyimagecuda_studio import LoadProject, set_node_parameter, run

with LoadProject("template.pics"):
    set_node_parameter("Text", "text", "Hello World")
    run("output.png")

Batch Processing

from pyimagecuda_studio import LoadProject, set_node_parameter, run

with LoadProject("certificate.pics"):
    for name in ["Alice", "Bob", "Charlie"]:
        set_node_parameter("Text", "text", f"Certificate for {name}")
        run(f"certs/{name}.png")

Global Variables

Control multiple parameters at once:

from pyimagecuda_studio import LoadProject, set_variable, run

with LoadProject("banner.pics"):
    set_variable("primary_color", (0.2, 0.4, 0.8, 1.0))
    set_variable("title", "Summer Sale")
    run("output.png")

Real-World: CSV Data

import csv
from pyimagecuda_studio import LoadProject, set_variable, run

with LoadProject("product_card.pics"):
    with open('products.csv') as f:
        for row in csv.DictReader(f):
            set_variable("product_name", row['name'])
            set_variable("price", f"${row['price']}")
            set_node_parameter("Image", "path", row['image_path'])
            run(f"cards/{row['id']}.png")

API Functions

# Load project (context manager handles cleanup)
with LoadProject(filepath, trust_code=False):
    
    # Modify nodes directly
    set_node_parameter(node_name, param_name, value)
    
    # Modify global variables
    set_variable(name, value)
    
    # Generate output
    run(output_path, quality=None)
    
    # Query project
    get_nodes()           # -> list[str]
    get_variables()       # -> list[str]
    get_node_parameters(node_name)  # -> dict

Security Note

Projects with Dynamic Text or Conditional nodes execute Python code. Set trust_code=True only for projects you trust:

with LoadProject("untrusted.pics", trust_code=True):  # ⚠️ Only if you trust the source
    run("output.png")

---

Roadmap to v1.0

Current: v0.1.0 Beta

Core functionality complete and stable.

Path to v1.0 Stable

• Bug fixes — Address issues reported by community
• Undo/redo — Full history system for GUI operations
• Memory pooling — Intelligent VRAM reuse to reduce memory footprint
• Expanded library — More nodes, effects, and generators

Timeline: v1.0 release when all critical features are stable and well-tested.

---

Contributing

Beta Release — Contributions, feedback, and collaborators welcome!

I'm very open to:

• Code contributions — Pull requests for bug fixes, features, or optimizations
• Ideas and suggestions — Feature requests and use case proposals
• Use case adaptations — I'm happy to adapt the program for personal workflows if they have general applicability

All contributions, questions, and feature requests go through GitHub Issues.

I'll respond as quickly as I can.

When reporting bugs, include:

• OS and GPU model
• Python version
• Error message or unexpected behavior
• .pics project file (if applicable)

---

License

MIT License

Copyright (c) 2025 Beltrán Offerrall Selma

See LICENSE file for details.

---