27 reactions
I really like the Bing Image Creator because it uses DALLE-3 to create images for free. If I had a premium account on Chat-GPT I would probably use that iteration of DALLE-3 - but I don't. I've become interested in testing the limits and tendencies of the Bing Image Creator, and so I have established a testing procedure with as few extraneous variables as possible. At first, I considered using a language-based prompt to create images, but I decided that there would be too much left up to interpretation by the LLM in regard to meanings of words. I have also come to realize that Chat-GPT 4 (which is the backbone for Bing Chat) will usually interpret a string of words differently (at least slightly) every time you present it with one. I decided to use images as the control/prompt, instead; specifically, images that I have generated with another AI: Craiyon.com which uses DALLE-mini to produce images.
This is my process:
I pick an image that I have generated on Craiyon that doesn't have any human faces in it (Bing Image Creator principally blurs all features of an image you present it with that it thinks are aspects of the human face, which I think is bullshit, especially for original images). Then I move to access the Bing Image Creator via the Microsoft CoPilot that is native to Windows 11/accessible through the Edge browser/available through the Bing Chat interface. It has three modes: "Creative", "Balanced", and "Precise" (the specific differences between these modes still elude me..). I proceed to send the "generate_graphic_image:" command to Creative mode attached to the control image from Craiyon. I do this three times, and each time, the interface returns 4 unique images, for a total of 12 images. I do this same thing for the Balanced and Precise modes and then I choose the 3 most interesting images from each set of 12. I am left with a control image, and 9 relatively high-quality images (3, each, from each of the CoPilot modes).
At best - with enough data - I can compare these sets and derive some level of expectation for the kinds of images that are generated by Bing within each mode of the interface.
At worst - I have a set of visually-appealing images that create an interesting variation on a theme.
Let me know what you think about this (and the image set attached to the post) in the comments!
n.0 is the Craiyon control
n.1 - n.3 are the 'Creative Mode' variations
n.4 - n.6 are the 'Balanced Mode' variations
n.7 - n.9 are the 'Precise Mode' variations
#BingImageCreator #DALLE3 #Craiyon #ImageGeneration #CoPilotModes #AIExperiment #VariationOnATheme
This is my process:
I pick an image that I have generated on Craiyon that doesn't have any human faces in it (Bing Image Creator principally blurs all features of an image you present it with that it thinks are aspects of the human face, which I think is bullshit, especially for original images). Then I move to access the Bing Image Creator via the Microsoft CoPilot that is native to Windows 11/accessible through the Edge browser/available through the Bing Chat interface. It has three modes: "Creative", "Balanced", and "Precise" (the specific differences between these modes still elude me..). I proceed to send the "generate_graphic_image:" command to Creative mode attached to the control image from Craiyon. I do this three times, and each time, the interface returns 4 unique images, for a total of 12 images. I do this same thing for the Balanced and Precise modes and then I choose the 3 most interesting images from each set of 12. I am left with a control image, and 9 relatively high-quality images (3, each, from each of the CoPilot modes).
At best - with enough data - I can compare these sets and derive some level of expectation for the kinds of images that are generated by Bing within each mode of the interface.
At worst - I have a set of visually-appealing images that create an interesting variation on a theme.
Let me know what you think about this (and the image set attached to the post) in the comments!
n.0 is the Craiyon control
n.1 - n.3 are the 'Creative Mode' variations
n.4 - n.6 are the 'Balanced Mode' variations
n.7 - n.9 are the 'Precise Mode' variations
#BingImageCreator #DALLE3 #Craiyon #ImageGeneration #CoPilotModes #AIExperiment #VariationOnATheme
| 528,000 mlx | |
| 197 | |
| 27 |
