Fer On Abstract Art

Working with Fer On Abstract Art

Fer On Abstract Art is a generative art technique that relies heavily on procedural layering and stochastic color placement. It's not a single software package or a trademarked method. People use the term to describe a workflow where you establish a rigid geometric or algorithmic skeleton first, then drive the visual complexity through overlapping transparent layers, noise functions, and controlled randomness. The results tend to look like mid-century constructivist painting mixed with modern shader art. I've been running this workflow for about four years now, mostly in Processing and TouchDesigner, and a lot of the frustration comes from treating it like a painting instead of a system. You set parameters, not brushstrokes. That shift in mindset takes some getting used to.

Fer On Abstract Art

Here is how I actually set up the workflow when I start a new piece, because the order matters more than people realize. First, you define your canvas geometry and grid structure. I usually start with a 4K canvas at 60 DPI. Then I build a base mesh or rule set — Voronoi cells, flow fields, or simple sine-wave interference patterns work fine. This base layer stays invisible in the final output most of the time. Its job is to dictate where everything else will sit. After the base structure is locked in, you introduce the color engine. This is where most people mess up. They grab a gradient preset and call it a day. Instead, assign each structural element a base hue from a restricted palette — five to seven colors maximum. Anything more and the image starts looking like a stock wallpaper generator output. I keep a running list of palette constraints in a JSON file so I can reproduce them across projects.

The actual "Fer On" part happens during the layering phase. You render each structural element as a separate pass, apply per-pass opacity values between 0.03 and 0.18, and composite them in sequence. The transparency stacking is what creates depth. A single element rendered at 0.15 opacity over another at 0.08 will produce a third color that never exists in your original palette. That emergent color mixing is the whole point of the technique. For noise and texture, I run simplex noise through a displacement map at roughly 2.3 frequency with 0.6 amplitude on the structural grid before rendering. This softens the edges just enough that the output doesn't look computationally sterile. You want it to feel handmade without losing the precision underneath. One edge case that caught me off guard: when you export your composited layers as PNG sequences for final rendering, alpha blending behaves differently depending on whether you're using pre-multiplied or straight alpha. I spent three days debugging a color shift that turned out to be simply the wrong alpha mode in After Effects. The workaround was switching my export pipeline to pre-multiplied alpha and re-rendering the composite in 16-bit EXR instead of 8-bit PNG. The color fidelity improved noticeably, especially in the midtone overlap regions.

Another practical note about color theory here — you should not trust your monitor calibration alone. Abstract layering compounds small color errors across dozens of passes. I run an X-Rite i1Display Pro calibration before each session and save the profile. Without that step, your final export will look completely different on screen versus print, sometimes shifting the entire tonal balance by several hundred Kelvins. Here are the tools I use and why: Processing for the procedural generation and base mesh work. It's free, it runs on anything, and the Java-based sketch system makes iteration fast. A typical composition takes between 45 minutes and 2 hours to generate the initial layers, depending on complexity.

Blender for any 3D structural elements. I use it mostly for simple volumetric forms that get flattened into 2D space afterward. The Cycles renderer handles transparency stacking better than Eevee, which clips alpha values at certain angles. After Effects or Nuke for final compositing. Both handle multi-layer alpha blending correctly. I prefer Nuke for this because the node-based workflow forces you to keep track of your layer order explicitly, and forgetting layer order in a node graph is much harder than in a timeline-based editor. If you're just starting out, do not jump into full composited renders right away. Run a test with three layers maximum. Verify that your opacity values are producing the color mixing you expect before you commit to a 40-layer composition. I've lost renders to crashed GPUs after 30 layers because a single opacity value was miscalculated early on and compounded across the stack.

One more thing people overlook: the export resolution and compression settings matter more than you'd think for this technique. If you're outputting to JPEG even at maximum quality, you introduce banding in the smooth gradient regions that transparency stacking creates. Always export to PNG or EXR at the very least. I use 16-bit EXR for the master files and downsample to 8-bit PNG only for presentation exports. The limitation worth acknowledging is that Fer On Abstract Art as a workflow does not scale well to real-time interaction. If you want this to respond to input live, you need to pre-bake most of the generative logic and cache the results. Running the full procedural pipeline in real time on consumer hardware usually caps out at around 30 frames per second with a moderately complex composition, which is fine for a gallery loop but useless for interactive installations. For real-time applications, the workaround is to pre-generate a library of frame variants — maybe 200 to 500 distinct compositions — and crossfade between them based on input. This gives the appearance of live generation while staying performant. It's less elegant but it's what actually works in practice.

File resources and code examples are scattered across GitHub and personal blogs. There is no single canonical source, which is one of the reasons the workflow feels undocumented despite being widely used. Search for "procedural layering generative art" or "stochastic color composition processing" and you will find enough material to get started within a few hours of reading.