3 ms·

I wouldn't at all agree that most Dall-E output is great out of the box. It has areas that it's good at and areas it's poor at. Here's a result for the prompt

I wouldn't at all agree that most Dall-E output is great out of the box. It has areas that it's good at and areas it's poor at. Here's a result for the prompt of "Woman with green skin, leaves instead of hair, wearing a simple dress, far shot, digital art, hyper-realistic, 8k, ultrahd," for example (all four images) https://imgur.com/a/f4d8N0u https://imgur.com/a/f4d8N0u You will note that none of them are even basically fulfilling the prompt, as well as all four being, in my estimation, ugly and uninteresting. That's not unusual for prompts that involve some element of the fantastic -- though there are corners of less-realistic digital art that it does do well.

TillE 4y ago
It's not the whole problem, but it looks like the "digital art" bit is dominating the style in a way you maybe didn't intend.
- aetherson 4y ago
  Here is the same prompt minus "digital art." https://imgur.com/a/aVbSxHe https://imgur.com/a/aVbSxHe You will still note that none of them are far shots, that no depicted character actually has fully green skin, and one of the four has nothing even remotely like leaves for hair. I mean, is it better? Sure. They're less ugly, though none of them are what I'd call great results. But they also aren't really doing a basically competent job of fulfilling the prompt, much less producing a particularly striking or interesting images. And my point is, outside of a few areas, this is what you get from Dall-E. Lots of misses, and if you're willing to put time into it and work on your results, a few hits. Don't get me wrong, I've gotten stuff from Dall-E that I think is great (I really like this "watercolor painting" for example: https://labs.openai.com/s/AQ7Wy5VHBWcLL5bJ5LbU5SuW https://labs.openai.com/s/AQ7Wy5VHBWcLL5bJ5LbU5SuW) but I think it misrepresents Dall-E to suggest that most of the time it produces basically good images. I'd say more like, "If you put time and attention into learning its quirks, in its best areas, it'll produce like one in ten images that are basically good." And, I mean, on some level that's incredible. You can produce 10 images in about three minutes in Dall-E and get some great stuff. But I think people mostly see the top 10% of what Dall-E produces.
  - yunwal 4y ago
    I think Dalle's ability to produce good images out of the gate is pretty limited, but I've found that using the fill-in feature along with existing images from google and photoshop, I can pretty much get anything I conceptualize with about 20 minutes of work and like 10 prompts. It's not fully removing humans from the equation, but you can take something that used to take days and make it a 20 minute operation.
    - aetherson 4y ago
      Yeah, the fill-in feature is great.

TillE 4y ago

It's not the whole problem, but it looks like the "digital art" bit is dominating the style in a way you maybe didn't intend.

aetherson 4y ago
Here is the same prompt minus "digital art." https://imgur.com/a/aVbSxHe https://imgur.com/a/aVbSxHe You will still note that none of them are far shots, that no depicted character actually has fully green skin, and one of the four has nothing even remotely like leaves for hair. I mean, is it better? Sure. They're less ugly, though none of them are what I'd call great results. But they also aren't really doing a basically competent job of fulfilling the prompt, much less producing a particularly striking or interesting images. And my point is, outside of a few areas, this is what you get from Dall-E. Lots of misses, and if you're willing to put time into it and work on your results, a few hits. Don't get me wrong, I've gotten stuff from Dall-E that I think is great (I really like this "watercolor painting" for example: https://labs.openai.com/s/AQ7Wy5VHBWcLL5bJ5LbU5SuW https://labs.openai.com/s/AQ7Wy5VHBWcLL5bJ5LbU5SuW) but I think it misrepresents Dall-E to suggest that most of the time it produces basically good images. I'd say more like, "If you put time and attention into learning its quirks, in its best areas, it'll produce like one in ten images that are basically good." And, I mean, on some level that's incredible. You can produce 10 images in about three minutes in Dall-E and get some great stuff. But I think people mostly see the top 10% of what Dall-E produces.
- yunwal 4y ago
  I think Dalle's ability to produce good images out of the gate is pretty limited, but I've found that using the fill-in feature along with existing images from google and photoshop, I can pretty much get anything I conceptualize with about 20 minutes of work and like 10 prompts. It's not fully removing humans from the equation, but you can take something that used to take days and make it a 20 minute operation.
  - aetherson 4y ago
    Yeah, the fill-in feature is great.