Because I’ve had the most success generating images with Google Gemini, I decided to try to get it to generate something like the design I drew for the character of “Queen Victoria” in the show Shipwrecked! An Entertainment.
First I asked it if I uploaded four sketches, could it generate a fifth one of Queen Victoria in mourning?
It replied:
I understand you'd like me to generate a fifth costume design rendering of Queen Victoria in a black dress, in the same style as four others you'd provide.
Unfortunately, I cannot directly "see" or "ingest" images you upload, nor can I replicate a specific artistic style from those images to generate a new one. My capabilities are primarily text-based.
To help you, you could try describing the style of your four renderings in detail. For example, you could tell me about:
The artistic medium: Are they sketches, paintings, digital art, etc.?
The level of detail: Are they highly realistic, stylized, abstract?
The color palette: Are they vibrant, muted, monochromatic?
The overall mood or aesthetic: Are they whimsical, dramatic, elegant, minimalist?
Any specific techniques or characteristics: Do they feature particular brushstrokes, textures, or linework?
With a detailed description of the style, I might be able to generate a textual description of a fifth rendering that aligns with your vision. However, I wouldn't be able to produce the visual image itself.
So I answered with all the parameters that it wanted to know to generate a description, it did so, and then I asked it to generate an image of that description. Despite having said it would not be able to produce an image, here’s what it created:
.
Not bad, given here’s my original sketch:
It’s worth mentioning that in the show, there are only three performers and this role was played by the one who was a 6’2” tall man on his knees (hence the bustle style to hide his legs) and he had a 30-second quick change into it. Throughout the series of renderings, I drew consistent features for each of the players when rendering costumes for the many roles they played.
But those are factors that make it unlikely that generative AI will become agile enough and manipulable enough to be helpful in the field of costume design in the development of final designs. But the future may prove me wrong. As an initial sketch illustrating how I envisioned the character of “Queen Victoria,” setting aside the fact that we would have to figure out how the part would be played by the 6’2” actor, the Gemini sketch is acceptable.
Just on a lark, I asked Gemini to generate another version of a Queen Victoria costume design rendering in the style of Erté & got this:
Meanwhile, Microsoft Copilot struggled with this because Queen Victoria is a real historical figure, an actual human who existed, and the license code of ethics prevented it from generating an image specifically mentioning her in the prompt, so I revised the language, asking for a sketch of a Victorian queen in mourning attire, and got these:
. .
Every time I think I have figured out something about how this technology works, I get an unexpected and surprising result causing me to question everything I thought I knew. In general, I feel like I’m past the point of diminishing returns with respect to image generating applicability in my discipline. The potentiality for clerical helpfulness with large language models feels more viable. .
Comments
Post a Comment