AI Watercolor Tool Set: I Tried Recreating a Plateau Mont-Royal Alley Exactly

One afternoon in late September, I was standing again in that alley in the Plateau. The narrow lane you reach by turning off Rue Drolet, between houses with their exterior brickwork half peeled away. On a wrought-iron outdoor staircase, a flowerpot someone had left behind was catching the sunlight and casting a long shadow. I had painted this scene three times over the past few weeks. To be precise, I fed the same reference photo and similar prompts into three different AI watercolor tools and compared the results.

Painting watercolors between motion design sessions is a long-standing habit of mine, and lately I've been using AI tools more often to generate rough sketches or colour references. Watercolor as a medium relies on bleeding and happy accidents, which also makes it one of the hardest styles for AI to imitate. So I decided to do a direct comparison.

Test Conditions

I fed all three tools the same photo (a backlit alley shot taken on my phone) and gave them "watercolor, soft edges, warm afternoon light" as the key prompt. I'll refer to the tools simply as A, B, and C. Each ran between 15 and 40 Canadian dollars for a monthly subscription, and the free trial credits were more than enough for testing.

Texture and Bleeding

A produced clean results. Too clean, in fact. The brick surfaces were rendered uniformly, as if a Photoshop filter had been applied, and there was none of that feeling you expect from watercolor, where pigment seeps along the grain of the paper. B, on the other hand, aggressively generated bleeding effects, and the textures resembling water pooling marks on paper were fairly convincing. The issue was that even parts that should have been straight lines, like the staircase railing, bled and ran, leaving the overall forms mushy. C was the most impressive in this regard. It distinguished between rough treatment for the coarse brick surfaces and smooth treatment for the sky, and its paper texture simulation was the only one among the three that looked natural.

Light, and Its Warmth

The afternoon light in Plateau Mont-Royal alleys is distinctive. The buildings are low enough that light reaches deep into the alleys, and as it bounces off the old red brick, the very air takes on a warm hue. None of the three tools captured this perfectly. A got the direction of the light right but the colour temperature was too cool. B pushed everything into an excessive orange cast, making it look like a sunset photo. C fell somewhere in between. It was a nice touch that it added the subtle violet of reflected light within the shadows, but the highlights were blown out to pure white.

When I actually paint watercolors, I leave the bright areas as bare white paper. Not applying paint is what creates the light. The AI tools didn't seem to grasp this idea of leaving space empty. All three tried to fill in the bright areas with at least a faint wash of colour.

The Human Touch

The biggest difference wasn't in technical accuracy but in atmosphere. In my own watercolor of the same alley, there's a stain under the staircase shadow where I used too much water. That stain is exactly what draws the eye there, and it brings back the memory of a cat curled up in that spot. The AI results have no such accidents. Every instance of bleeding is calculated, and the beauty of the unintended is missing.

I think C is the one I'll reach for most often. Its texture handling is good enough as a starting point, and it works well for pulling out colour palettes as reference. That said, none of the three are yet at a level where I could use the output as a finished piece. Photos of that alley keep accumulating on my phone with each passing season. Next time, I'm thinking of running the snow-covered winter version through them. I'm curious whether they can imitate the feeling of paint freezing on paper.