NICOLA GINZLER

Senior Graphic Designer
Search
Close this search box.

Fun with AI Font Friday 1 – blending images

OK, I’m still playing with Midjourney AI image generation software, but this time using my own images to avoid copyright infringement. This round actually involves some fonts, or rather, Midjourney’s attempts at type. None of the image generation software I’ve found is currently good at reading, understanding or reproducing text, although it may (but doesn’t always) recognize it as such.

How this all started

Midjourney has a /blend command which combines two to five images. I used two of my own fine art pieces, below, “Nicola Likes To Be Warm” and “Kiss (Nicola Loves You When You Leave).” I’ve had the images for a while and I wanted to see what would happen if I blended similar but not identical images, even though (as far as I can tell) that’s not what the command is for.

Image 1. Illustration with flat areas of color. Red background; female figure in short black sleeveless dress, black tights and shoes is holding a gasoline can in her right hand and a lit match in her left. Black text down right-hand side reads "Nicola / likes / to be / warm."
Image 1 used with Midjourney /blend command
Image 2. Illustration with flat areas of color. Red background; female figure in short black sleeveless dress has eyes shut and is blowing a kiss. She has stylized fringed eyelashes and her lips are bright red. Her fingernails are short and painted red.
Image 2 used with Midjourney /blend command
Image 3. Illustration with flat areas of color. SAME AS IMAGE 1, BUT FLOPPED LEFT TO RIGHT. Red background; female figure in short black sleeveless dress, black tights and shoes is holding a gasoline can in her right hand and a lit match in her left. Black text down right-hand side reads "Nicola / likes / to be / warm."
Image 3 used with Midjourney /blend command (flopped version of Image 1)

First I blended Images 1 and 2 only.

For the blended images, Midjourney kept the black and white of Image 1, but had a darker red and mostly used the composition of Image 2. It did make an attempt at type, to match Image 1, and showed the figures to below the waist.

Image 4 (blended) with flat areas of color. Red background; female figure in short black sleeveless dress, shown to below waist, has eyes shut and sad expression. She has stylized fringed eyelashes and her lips are bright red. Her hand has six fingers. A red glove floats sideways in front of her. Her left arm ends at the elbow. Text on right side reads "iimd vvoal."
Image 4 (blended)

Midjourney’s most successful blended image.

Wins:

  • The face has symmetrical, attractive features and a neutral expression.
  • The body has a similar posture to that of Image 2.
  • The body is shown to below the waist, a combination of Images 1 and 2.
  • The stylized eyelashes were matched.
  • Black text on the right side.

Fails:

  • The left hand has six fingers and isn’t held up for blowing a kiss.
  • The red fingernails have been replaced by a floating red glove.
  • The left arm ends at the elbow.
Image 5 (blended) with flat areas of color. Red background; female figure in short black sleeveless dress, shown to below waist, has eyes shut and pained expression. Her lips are bright red. The left hand and red fingernails have been replaced by a floating red mass. The right hand is distorted and has elongated black ovals replacing the fingernails. Text on right side reads "lilo ccr wol."
Image 5 (blended)

Somewhat less successful.

Wins:

  • The colors match Images 1 and 2.
  • The body has a similar posture to that of Image 2.
  • The body is shown to below the waist, a combination of Images 1 and 2.

Fails:

  • The face has less symmetrical features and a pained (non-neutral) expression.
  • The left hand and red fingernails have been replaced by a floating red mass.
  • The right hand is distorted and has elongated black ovals replacing the fingernails.

Then I blended Images 1, 2 and 3, to give more weight to the standing figure.

Midjourney won’t let you upload the same image twice so I flopped Image 1 to produce Image 3. Here’s what I got.

Image 6 (2x2 blended images) with flat areas of color. Red backgrounds; female figures in short black sleeveless dress, shown to below waist, have eyes shut and neutral or pained expressions. Their lips are bright red. On each, one hand is raised and there is black or white text on the right or left side.
Image 6 (4 blended images)

More distorted.

Wins:

  • The colors match Images 1, 2 and 3.
  • The bodies have a similar posture to that of Image 2.
  • The bodies are shown to below the waist, a combination of Images 1, 2 and 3.

Fails:

  • The faces have less symmetrical features and some have a pained (non-neutral) expression.
  • On each, at least one hand is raised, but none matches Image 2.
  • There is black or white text on the right or left side.
Image 7 (2x2 blended images) with flat areas of color. Red backgrounds; 4 female figures in short black sleeveless dress, shown to below hip, have eyes shut and neutral expressions. Their lips are bright red. On each, both hands are at waist height or higher, holding an object. There is writing on the dress or object.
Image 7 (blended images)

Less distorted but with elements added.

Similar wins:

  • The colors match Images 1, 2 and 3.
  • The bodies have a similar posture to that of Image 2.
  • The bodies are shown to below the waist, a combination of Images 1, 2 and 3.
  • The faces have symmetrical features and neutral expressions.
  • One figure has text to the right side.

Fails:

  • Suddenly, the figures are holding random objects.
  • There is red or white text on the figure or the object in the hand.

Conclusions

  1. Midjourney is OK but not great at blending images.
  2. Midjourney recognizes the faces and hands in my illustrations, but is less good at rendering them. Midjourney has trouble with faces and hands in general, but even more so when the images are not photographic.
  3. Midjourney suuuuucks at rendering type. We knew this already but this exercise is just more more confirmation.

More Midjourney attempts below

Image 8 (2x2 blended images) with flat areas of color. Red backgrounds; 4 female figures in short black sleeveless dress, shown to below hip, have eyes shut and neutral expressions. Their lips are bright red. On each, both hands are at waist height or higher, holding an object. There is writing on the dress or object.
Image 8 (blended images)
Image 9 (2x2 blended images) with flat areas of color. Red backgrounds; 4 female figures in short black sleeveless dress, shown to below hip, have eyes shut and neutral expressions. Their lips are bright red. On each, both hands are at waist height or higher, holding an object. There is writing on the dress or object.
Image 9 (blended images)

The owner of this website has made a commitment to accessibility and inclusion, please report any problems that you encounter using the contact form on this website. This site uses the WP ADA Compliance Check plugin to enhance accessibility.