I've been researching agentic AI workflows a bit, and my mind just automatically maps it all to professional creative workflows.
And, I started thinking about a use case where orchestrated agents are instructed to create brushstrokes and automatically layer the brushstrokes one by one emulating a digital illustrator or painter. I don’t think it could hit the quality level, but would it try? I’m curious how far down that path it could go.
Is it possible to instruct an AI agent to orchestrate multiple AI agents to create brushstrokes, and auto paint in brushstrokes and layers and have it auto refine until high quality?
I came up with a hypothethis (actually it's more of thought experiment at the moment) that I'm curious to prove or dis prove.
Could you prompt Manus.im, or structure an ai agent to create layer by layer brushstrokes like auto-brush art?
Hypothesis is that it may be able to paint digitally rather than use an ai model. It would either achieve it somehow, probably by creating agents to emulate digital art brush painting tools, or it wouldn't, or achieve a third outcome I haven’t seen..
UPDATE:
I finally got an invite to Manus, so I ran part of my informal test. I asked Manus to create art brushstroke by brushstroke like digital art by an artist and not by using standard AI image generator tools.
I was thinking that if it could do that, that adds another criteria layer to the question of what is considered significant human manipulation in ai art ownership.
The idea of using Manus to simulate brushstroke-by-brushstroke creation touches directly on a fundamental legal and philosophical boundary in AI art:
If AI replicates the human process of creation, stroke by stroke, is the human prompting and guiding it enough to constitute authorship—or does the process have to involve actual input at that level of manipulation?
New Criteria Layer in Copyright?
If Manus can be prompted to paint in layers like a real human would — say:
underpainting
blocking
detailing
selective stylization
... then is the prompt writer performing a creative act of sufficient originality — similar to how a film director isn’t holding the camera or painting the matte but is still the author of the final vision?
Manus created algorithms for the "brush strokes" inside the Manus code sandbox and it gve me these images along with each reusable algorithm.
The quality is not as important here as the process it decided to create - and the repeatable code it created.
In theory, I could level up the brushstroke algorithms, or I think the next test will be in the Google AI Studio where you can have Gemini interact with your computer. I would love to see what it does with all my Illlustration tools.
That test is still on my list.
The test so far shows that manus can decide to create algorithms for drawing brushstrokes to simulate drawing digitally. I should see if the code is what it’s saying it is. The brushstrokes are flat, and the quality overall resembles outputs like gpt svgs. That makes sense given the python tools they both invoked.
(Images created by algorithm-based “brushstrokes” in Manus sandbox.)
