By TextureLoRALab (Shitsukan)
2026: The Year Art Got Tired of "Flat"
Something shifted in the AI art market this year. And it was not subtle.
Gallery owners report collectors demanding touchable texture. Interior designers now specify "no flat digital prints." Art fairs are seeing a surge in impasto, encaustic, mixed media — anything with physical surface presence.
The industry calls it the "Anti-Algorithm" trend. A broad consumer rejection of the smooth, perfect, frictionless aesthetic that characterized the first wave of AI-generated imagery.
To borrow a phrase from one art market forecast: enough of fake, fast, and flat.
What "Anti-Algorithm" Actually Means
It is not anti-AI. This distinction matters.
The pushback targets not AI as a tool, but the default output of AI as a tool. The smoothness. The eerie perfection. Every surface looking "rendered" rather than "made." (I call this the Wet Problem — everything looks like it has been lightly misted.)
What people are looking for:
Traces of process. Brushstrokes, tool marks, drips, irregularities — evidence that a chain of physical decisions took place.
Material honesty. Surfaces that look made of something specific. Not "generic digital medium."
Tactile invitation. Images that make you want to touch them.
In other words, people want materiality. Work that feels physically present even when viewed through a screen. (For a deeper discussion of why AI images lack this, see Why AI Doesn't Know the Weight of Paint.)
The Opportunity for AI Artists
The Anti-Algorithm trend does not mean AI art is over. It means generic AI art is over. The bar has moved from "Can AI make beautiful images?" (already solved) to "Can AI make images that feel real — with the weight and texture of physical objects?"
I believe this is a solvable problem. And LoRA is the key. The challenge breaks down into three layers:
Layer 1: Surface Appearance (Partially Solved)
Good LoRAs for brushstroke styles, canvas overlays, and painterly effects already exist. ClassipeintXL, OilpaintZ, impasto LoRAs — excellent tools for visual texture. But appearance alone is not enough. Photoshop's canvas texture filter also gives you appearance. Anti-Algorithm audiences can tell the difference.
Layer 2: Material Behavior (Emerging)
This is the frontier. Not "what does oil painting look like" but "how does oil paint behave on linen canvas." I analyzed this distinction in detail in Style LoRA vs Texture LoRA. The core point: material behavior LoRAs learn the physics of a specific material, not the aesthetics of a specific style. Logic lives in the surface. Cracks follow stress patterns. Gold does not just shine — it fractures along fault lines.
Layer 3: Scale Consistency (Untouched Frontier)
The hardest challenge. Truly convincing physical texture must work at macro, mid-range, and full-view scales. The 3-Distance Method is designed specifically to address this — custom training image sets photographed at three scales, so the model learns consistent behavior from close-up grain to environmental context. Most current LoRAs work at one, maybe two distances.
Commercial Implications
Here is the part that matters for anyone generating commercially.
The market for "AI art prints" is saturated. You can buy AI-generated wall art on Etsy for $15. Because the output all looks the same — smooth, stylized, interchangeable — margins keep shrinking.
But the market for textured, material-specific AI art — output that looks like a photograph of a real artwork — is wide open. Almost nobody is generating it yet.
Think from the buyer's perspective:
Generic AI art print: "Nice image." → $15
AI-generated image with convincing gold leaf texture, visible material behavior, print-resolution detail: "Wait, is this a photo of a real painting?" → Completely different price point.
The weight of a model's [Commercial Use OK] label increases sharply when output quality crosses the threshold from "digital print" to "convincing reproduction."
The Bigger Picture
The Anti-Algorithm trend might actually be good news for AI artists. Higher standards mean low-effort noise gets filtered out, creating space for those who care about craft.
If you are already working on generation with texture, materiality, and physical presence in mind, the market shifts described here may be worth watching.
The future of AI art is not smoother, more photorealistic, more perfect output. It is output that carries traces of material reality — even if it was never physically made.
Physical texture is the new frontier. The algorithm is learning to be Anti-Algorithm.
Start Here
SHIFUKU Gold Leaf v1 — Physical Texture LoRA for SDXL (Commercial Use OK)
SHIFUKU Gold Leaf v2 — Three gold leaf techniques: kiribaku, sunago, noge (Commercial Use OK)
SHIFUKU Kintsugi — Physical Texture LoRA for SDXL (Commercial Use OK)
SHIFUKU Hamon Steel (Beta) — Japanese sword texture (Commercial Use OK)
Related:
Why AI Doesn't Know the Weight of Paint
Style LoRA vs Texture LoRA — They Solve Different Problems
Training Texture LoRAs from Real Materials: A 3-Distance Method