Jump to content

Will DALL-E take up creative jobs or make them more efficient?


spiritual_lama

Recommended Posts

Speaking as someone who buys a lot of art, and has seen a lot of DALL-E examples; I don't see me switching to entirely using AI any time soon. Right now it's good for prototyping/concepts and placeholders for market testing/pitches. A lot of commercial-release art requires specific details dictated by things like consumer psychology. I expect that in the near future there will be people whose job is to "engineer" DALL-E art; people who have a strong intuitive sense of what inputs will get the desired results (like how IT nowadays is more about knowing what to google). There's also a question of what kind of inputs, or how many reference images, it would need to be supplied to be able to replicate a new/original character.  

To me, DALL-E is a jacquard loom. Probably it will reach a point where people who do their job the old way will get out-competed, But, if someone gains the skills needed to operate the tool in addition to their old skills, they get to be at the forefront of production revolution and reap the rewards. I've been in the game industry since 2001, and have seen a wide range of similar advancements. Every advancement was followed by more developers entering the industry and more games coming out. They made the markets more competitive as the bar to entry was lowered so that people with less time and money can get in on it. 

Like, I expect the near-future workflow looks like "Traditional Artist is assigned new character development. Traditional Artist uses DALL-E to churn out a variety of design options, and passes those back to the Producer. The Producer selects a DALL-E work as a basis and provides notes with that framework. The Traditional Artist then makes a proper piece based on the annotated DALL-E piece." If I had my own access to DALL-E, I could see me using it to create annotated reference images for what I send to Fiverr artists.

  • Like 6
  • Up 1
Link to comment
Share on other sites

For now I can only see dall-e being used in decorative elements like on wall etc... For people wanted cool looking things. As for game, comic, film industry it doesn't change anything. There were already similar tools like dall-e for decade for concept artists it's "photoshop". Concept artists bash images more deliberately could probably chunk out more useful images or prototypes. I could probably see dall-e used for testing out color schemes.

  • Like 6
  • Up 1
Link to comment
Share on other sites

I'm not an expert. I haven't tried DALL-E or DALL-E2. Though DALL-E2 looks impressive from videos about it. Though I think they all can be inaccurate. eg. on faces/hands.

I think DALL-E2 costs to use currently and it may not be open to everyone (I think you need to be invited). I think they'll integrate more things like that over time into software (like photo editing software), so some things may become easier. If the software comes with that functionality and it's cheap enough and easy enough for many people to use it may cut down on some people asking some sellers to use it. But there are also questions about rights. Would the buyer have all necessary rights to use it commercially? What if someone else put in the same text and got the same results and used it?

Here's a few images from a free Stable Diffusion demo. It seems a lot lower quality than DALL-E2 (maybe especially for faces, though it gets hands wrong too) but it's free to access and use.

The images are only 512x512 pixels whereas DALL-E2 uses 1024x1024 pixel ones (though that still may be a bit low).

ai-2.png.a417cf6a03ecc0a06cf5d1005b819780.pngai-16.png.145aac13b2e3dc15c7eb993b4d69f6a0.pngai-17.png.8b36f6f2c4fbe967e614be0ac768334a.pngai-1.png.912d7b98166f49b30fbf3dcc93b7883d.pngai-10.png.fa1fcb88afcda63663164c2296d9e442.pngai-11.png.d938b23e4f8c7bbbc0a6890d53da7f3c.pngai-12.png.4839886919e7ce364f231fd7b096de5e.pngai-15.png.efd52972d475eb181a84dc8f287f585e.png

Also this Stable Diffusion one (and I assume others too) wouldn't be good if you needed text as the text you ask it to add goes wrong. eg. in the images above where it looks a bit like "Fiverr" it was actually supposed to say "Fiverr" and not the additional/changed text and the logo in the 3rd image was supposed to say "My Logo" not "ML". So it wouldn't be good for Fiverr gigs where it actually needed to show certain text (unless the seller added it manually themselves in their image editing software). Also revisions would also probably be difficult, eg. in this AI you couldn't take an image it created and then ask it to change it in a particular way.

So I think they will get better though in future. There is a gig that uses something like DALL-E for music videos which looks good quality. Maybe there will be more stuff like that in future.

Edited by uk1000
  • Like 5
  • Up 1
Link to comment
Share on other sites

16 hours ago, moikchap said:

I expect that in the near future there will be people whose job is to "engineer" DALL-E art; people who have a strong intuitive sense of what inputs will get the desired results (like how IT nowadays is more about knowing what to google)

This is indeed a very interesting take. Hadn’t thought of this before. Interesting 🤔

  • Like 2
Link to comment
Share on other sites

1 hour ago, uk1000 said:

I'm not an expert. I haven't tried DALL-E or DALL-E2. Though DALL-E2 looks impressive from videos about it. Though I think they all can be inaccurate. eg. on faces/hands.

I think DALL-E2 costs to use currently and it may not be open to everyone (I think you need to be invited). I think they'll integrate more things like that over time into software (like photo editing software), so some things may become easier. If the software comes with that functionality and it's cheap enough and easy enough for many people to use it may cut down on some people asking some sellers to use it. But there are also questions about rights. Would the buyer have all necessary rights to use it commercially? What if someone else put in the same text and got the same results and used it?

Here's a few images from a free Stable Diffusion demo. It seems a lot lower quality than DALL-E2 (maybe especially for faces, though it gets hands wrong too) but it's free to access and use.

The images are only 512x512 pixels whereas DALL-E2 uses 1024x1024 pixel ones (though that still may be a bit low).

ai-2.png.a417cf6a03ecc0a06cf5d1005b819780.pngai-16.png.145aac13b2e3dc15c7eb993b4d69f6a0.pngai-17.png.8b36f6f2c4fbe967e614be0ac768334a.pngai-1.png.912d7b98166f49b30fbf3dcc93b7883d.pngai-10.png.fa1fcb88afcda63663164c2296d9e442.pngai-11.png.d938b23e4f8c7bbbc0a6890d53da7f3c.pngai-12.png.4839886919e7ce364f231fd7b096de5e.pngai-15.png.efd52972d475eb181a84dc8f287f585e.png

Also this Stable Diffusion one (and I assume others too) wouldn't be good if you needed text as the text you ask it to add goes wrong. eg. in the images above where it looks a bit like "Fiverr" it was actually supposed to say "Fiverr" and not the additional/changed text and the logo in the 3rd image was supposed to say "My Logo" not "ML". So it wouldn't be good for Fiverr gigs where it actually needed to show certain text (unless the seller added it manually themselves in their image editing software). Also revisions would also probably be difficult, eg. in this AI you couldn't take an image it created and then ask it to change it in a particular way.

So I think they will get better though in future. There is a gig that uses something like DALL-E for music videos which looks good quality. Maybe there will be more stuff like that in future.

Wow stable diffusion looks pretty good for being free. Such softwares are really difficult to make. Future is indeed exciting!

  • Like 4
Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...