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).
Quick.mp4
Variables.and.more.mp4
- 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
- Simple Python API: Load
.picsprojects 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")- 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
- Python 3.10+
- NVIDIA GPU (GTX 900 series or newer)
- OS: Windows 10/11 or Linux
pip install pyimagecuda-studiopyimagecuda-studio
# or shorter alias:
pics- 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
- Rounded Corners — Smooth corner radius
- Drop Shadow — Blur, offset, color control
- Stroke — Inside/outside outline
- Vignette — Edge darkening
- Chroma Key — Background removal by color
- 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
- Brightness — Linear adjustment
- Contrast — Midpoint scaling
- Saturation — Color intensity
- Hue — Color rotation
- Gamma — Non-linear correction
- Opacity — Alpha blending
- Vibrance — Smart saturation boost
- 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
- Zoom — Zoom in/out with offset
- 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
- Random Out — Randomly select between inputs
- Conditional — Python-based routing logic
- Split — Duplicate to 2 outputs
Design templates in the GUI, automate generation with Python.
from pyimagecuda_studio import LoadProject, set_node_parameter, run
with LoadProject("template.pics"):
set_node_parameter("Text", "text", "Hello World")
run("output.png")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")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")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")# 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) # -> dictProjects 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")Core functionality complete and 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.
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
.picsproject file (if applicable)
MIT License
Copyright (c) 2025 Beltrán Offerrall Selma
See LICENSE file for details.