For the Runway demo: I prompted the photorealistic image was a combo of these words - "A savvy, older female executive facing camera in a neutral office environment" paired with a style reference image I'd generated weeks ago with the prompt - "A four-part corporate org chart in the style of Paul Rand." So, an older assignment became useful for this newer task. Now, let's clarify a little. For the Paul Rand style reference, I had generated 40+ images. And for the photorealistic older executive assignment, I tried out five different "in the style of" approaches, generating 20-30+ per style... to arrive at the solution I used. Lesson learned: It's about quantity long before it's about quality.
For the second image (the one Runway could detect a face in), it too is built on a text prompt - "2D illustration of an older female business executive facing camera in a neutral office, clear eyes and lips, smiling" with an image reference. That image reference was created by prompting - "A mural depicting corporate culture in transition in the style of Picasso."
Thanks for the thorough breakdown. I haven’t spent enough time getting good at the prompts, but this is a good example of how important the prompt is for the output. Thanks!!
This whole essay is really great, Tim. But, I'd be curious to see the prompts you used in midjourney to get the first two images.
For the Runway demo: I prompted the photorealistic image was a combo of these words - "A savvy, older female executive facing camera in a neutral office environment" paired with a style reference image I'd generated weeks ago with the prompt - "A four-part corporate org chart in the style of Paul Rand." So, an older assignment became useful for this newer task. Now, let's clarify a little. For the Paul Rand style reference, I had generated 40+ images. And for the photorealistic older executive assignment, I tried out five different "in the style of" approaches, generating 20-30+ per style... to arrive at the solution I used. Lesson learned: It's about quantity long before it's about quality.
For the second image (the one Runway could detect a face in), it too is built on a text prompt - "2D illustration of an older female business executive facing camera in a neutral office, clear eyes and lips, smiling" with an image reference. That image reference was created by prompting - "A mural depicting corporate culture in transition in the style of Picasso."
Hope this helps!
Thanks for the thorough breakdown. I haven’t spent enough time getting good at the prompts, but this is a good example of how important the prompt is for the output. Thanks!!