A blog related to original AI artworks. I make AI artworks using coined words and emoji as prompts.

How to Select Better Works Using ChatGPT

I provided the following artworks to ChatGPT-4o and evaluated them using several types of questions. In the end, the method of presenting multiple works and having it select the best appears to be effective for choosing higher-quality works. Below, I describe the three methods I tested.

Having It Explain Strengths and Weaknesses

When I asked, “Please explain this artwork’s strengths and weaknesses in about 10 lines,” it responded as follows (punctuation has been modified):

This artwork is notable for its abstract composition and vibrant colors. Its strengths include the bold use of primary colors—red, blue, and yellow—which give it visual impact. Black lines unify the piece, maintaining structural balance. Additionally, the overlapping colors and transparency create a sense of depth and dynamic impression.

However, weaknesses include an excess of elements that could cause visual confusion. The focus is unclear, and viewers may be unsure where to direct their attention. Organizing the elements more could result in a more refined piece.

It identified the weaknesses as “an excess of elements causing visual confusion” and “lack of focus,” suggesting “organizing the elements” as a solution. These observations apply to many other artworks as well, indicating a general shortcoming of AI art that may not have easy solutions for now.

Having ChatGPT Compare Multiple AI Artworks

Since differences between multiple AI artworks aren’t easy to distinguish with the above method, I tried having ChatGPT compare several works.

Among the artwork “nv2jn-7m” above and the two artworks shown here (ndyfgrq7, 19njbskq), which do you think has the highest level of completion?

The following response was obtained:

I will assess which of these three artworks seems the most complete from the perspectives of structural cohesion, color balance, visual depth, and movement.

• **Characteristics of “nv2jn-7m”**
  Structural Cohesion: Bold black lines give rhythm to the composition, and colorful blocks are effectively placed.
  Color Balance: The vibrant red, blue, and yellow create a visually impactful composition.
  Depth and Movement: Strong sense of three-dimensionality, with overlapping layers that convey depth and movement.
  
• **Characteristics of “ndyfgrq7”**
  Structural Cohesion: Finely arranged lines create an overall regular pattern, but the complexity can feel overwhelming, dispersing the gaze.
  Color Balance: The mix of monotones and colorful blocks feels visually heavy, and overall cohesion may be somewhat lacking.
  Depth and Movement: There is depth, but the lack of movement gives it a more static impression.
  
• **Characteristics of “19njbskq”**
  Structural Cohesion: Composed of regular lines emphasizing a Mondrian-style grid structure.
  Color Balance: The color distribution is balanced, with ample white space, creating a sense of openness.
  Depth and Movement: Layers are present, though three-dimensionality is weaker, giving a relatively flat appearance.
  
• **Overall Evaluation**
  Among the three, **“nv2jn-7m”** is considered the most complete. This piece has an effective rhythmic structure and dynamic layers, with excellent color balance. Its depth and movement create a visually compelling appeal.

Since I also rated “nv2jn-7m” highest, I can say that my evaluation aligned with ChatGPT-4o’s. This method may therefore be useful. However, ranking all works this way would require giving ChatGPT significant time.

Having It Compare with Famous Works

I also compared an AI artwork with a famous painting (*On White II* by Kandinsky).

When I asked, “Compare the above artwork *On White II* with this one (nv2jn-7m) and note the strengths of each in about 10 lines,” it responded as follows:

These two artworks, “On White II” and “nv2jn-7m,” both feature abstract compositions with striking color and shape contrasts but exhibit unique visual approaches and strengths.

• **Strengths of “On White II”**
  Fluid Composition: Lines and shapes flow throughout, creating movement and rhythm, giving it a dynamic visual effect.
  Combination of Diverse Shapes: Various shapes—lines, curves, circles—combine skillfully, providing a varied visual experience.
  Use of White Space: The white background accentuates shapes and colors, giving the piece a breathing space.
  Color Balance: Colors are well-balanced, creating a calm visual impression without any color dominating excessively.
  Hint of Symbolism: The geometric shapes feel symbolic, stimulating viewers’ imagination.
  
• **Strengths of “nv2jn-7m”**
  Strong Color Contrast: The bold contrast of black, red, blue, and yellow creates a powerful visual impact.
  Layered Composition: The overlapping layers emphasize depth, giving a three-dimensional impression.
  Sharp Lines and Shapes: The frequent use of straight lines and sharp shapes creates a sense of tension and precision.
  Dynamic Arrangement: The scattered shapes convey movement across the canvas, creating an energetic impression.
  Modern Design Elements: The graphic design elements create a sophisticated, modern feel.
  
• **Comparison Summary**
  “On White II” offers a soft, flowing composition with ample space, while “nv2jn-7m” stands out with bold colors and shapes, delivering a more intense and energetic impression. Both provide distinct visual experiences, each with unique appeal.

This question aimed to see if weaknesses ChatGPT had identified in AI artworks would also appear in famous artists’ works and whether they would be masked and judged more favorably. This answer seems to confirm that possibility.