Image is pictures. Effects such as Blur, Bitmosh and Depth Lens. Analysis that reads a frame: depth, detection, segmentation, tracking, optical flow. Generation from latents and diffusion. Every operator takes a TOP and returns one, and the work stays on the GPU between them.

The Voro column of this view holds Kernel FX, Script FX and ONNX FX. The shipped effects and models are those three with a source or a model already chosen, so anything you write sits in the same menu.

Start with First pipelines.

Modules

• Convert Turn one wire kind into another on purpose. Convert names the input class, the output class and the method, and never happens by itself.
• Kernel FX Write a GPU kernel in Python. Reload compiles it to native CUDA code, and the node runs it on every frame with no Python in the loop.
• Script FX Write a Python function that processes each frame. Use GPU tensors to change an image or run a Python model.
• ONNX FX Run an ONNX model on the GPU. Point the node at a model file and a config, press Reload, and the graph has a new model operator.
• Blur Blur an image along a motion field, around a centre, or in all directions. Use the Warp preset to displace it.
• Bitmosh Break a moving image into codec trails, drifting blocks and damaged slices, then repair the picture with a reset.
• Brushstrokes Build a painting from a moving image, with persistent brush marks, wet paint, pigment palettes and a textured canvas.
• Contour Extract and shape image edges into lines, nested rings, cutouts or a reusable matte.
• Darkroom Process an image as film, with exposure, development, print colour, grain and gate wear.
• Depth Lens Use a depth image to focus a picture, shape its bokeh and add lens colour fringes or flare.
• Ink Bleed Lay down ink and let water carry it through paper, fabric or another surface. Build continuous marks or separate impressions.
• Isochrome Turn image structure into coloured stress fringes seen through a pair of polarizers.
• Printer Turn an image into a printed sheet, with halftone plates, spot ink, paper texture and a controllable print rate.
• Teleloop Run an image through a simulated video signal, worn tape and CRT display, with an adjustable feedback loop.
• Upscale Enlarge an image with a compatible super-resolution model. The supplied Real-ESRGAN configuration scales each axis by four.
• Depth Anything Estimate relative depth from a single image and use it to drive focus, displacement or masks.
• Detect RF-DETR Find objects in an image and publish their boxes, classes, confidence scores and a field for effects.
• Segment SAM3 Find subjects from text concepts or spatial prompts, then publish their masks, outlines and detection data.
• Track Anything Follow a grid or a chosen set of points through video, with position, confidence and visibility outputs.
• Face Track Track a face's pose and expression, publish control channels, and drive an animated portrait.
• Optical Flow Estimate pixel motion between frames and use the vector field to steer blur, warping or other image effects.
• Video Depth Anything Estimate relative depth across a moving sequence with a model that retains information from earlier frames.
• StyleGAN Generate images from latent values using a compatible ONNX generator and its matching configuration.
• StreamDiffusion Transform a live image with a configured diffusion model, with optional control images, style input and feedback.
• Portrait Rig Animate a still portrait from Face Track motion, with baseline capture and separate expression, pose and translation controls.
• Latent Seed Create a repeatable 512-component latent from a seed, or form one from an incoming image.
• Latent Compose Combine one to four compatible latents with independent weights and a choice of mixing modes.
• Latent Transform Normalize, scale, offset, limit or modulate a 512-component latent before image generation.