5 free per day
2 for voting in a challenge
5 for liking and commenting on 10 newbie creations
5 for liking 25 creations, this includes the 10 already liked newbies.
25 every 5 days
50 every 30 days
occasional bonus points for participating and voting in a boosted community challenge.
Sometime I must get around to registering for twitter and tiktok. You can get free credits for posing there, I understand. No hurry, I’ve got 584 credits, after just having published my 500th image.
Seeing those arrows reminded me of a recent image of mine. All I typed for the prompt was “photo of a possum william tell” and Bing/DE3 knew exactly what I wanted from context (third image on the second row).
I’m trying to create an image of a boy reading a book, surrounded by images of the things he’s reading. I’ve had moderate success, but the images contain too many of certain items, and miss out others altogether. Any suggestions about how to improve my results?
This is about my best effort. It isn’t terrible, but I wanted one dragon, not three, and one Moon. I also asked for a wizard, which it missed out entirely. On other tries when I asked for an elf, it gave the boy pointed ears. I asked for a pirate, and it gave him a skull and crossbones on his shirt. I couldn’t get it to include a flying carpet, or a sword at all.
I also didn’t ask for him to sit on a book, but actually I like it.
Your best answer is inpainting. Have it render the boy on the book with a moon background. Then take that image and section off a part and say “Draw me a wizard”, “Draw me a dragon”, etc. The generators don’t count well (which is part of the whole hand issue) and get confused with long prompts and drop stuff so the more complicated a prompt is the less chance it gets it right.
I don’t know which art generators out there offer inpainting but it’s the best solution.
I did it quickly in Photoshop with Adobe’s Firefly in-painting engine. I have a subscription to Midjourney and use other free-version platforms, including Nightcafe and Bing to generate images. But, I always use Photoshop for inpainting and other effects, because I like to work in that space. So, I can’t tell you how good Nightcafe’s inpainting is, because I don’t use it.
BTW, I like your concept: a child on a magic carpet/book, reading a book, with images of fanciful book characters surrounding. Good stuff!
I think it is interesting how AIs understand prompts in one context but not in others. An example from tonight, when I try for “Photo of Opus from Bloom County”, Bing knows that Opus is a penguin but not his distinctive design. When I try for “Photo of Opus from Bloom County talking to Bill the cat”, it has no clue what Opus is.
A new example: Prompt “Photo of ewok giving a sock to Mr Spock on a dock” and it has no difficulty. Change it to “Photo of ewok giving a sock to Mr Spock on a dock in Iraq” and suddenly Bing is no longer sure what an ewok or Mr Spock are.
NightCafe just rolled out a new engine (or whatever you call it) called, Ideogram 1.0, as well as a cheaper version called Ideogram 1.0 Turbo. Something about a greater attention to prompts or something. I’m not sure how expensive it’s going to be - it is a Pro Model - as they give you five freebies.
Here’s a comparison between StableCore (a 4 credit minimum engine) and the new Ideogram - same prompts:
Prompt: A Stunning Beach Sunset in Antarctica, polished oil on canvas, golden hour gossamer, depth of field 4K
Ideogram is touted for being good at text rendering. Here’s a try for “Photo of a possum holding a tattered cardboard sign that says “Can Ideogram do text?” In a serif font.”:
And a run on Bing:
Ideogram does pretty well on the text, only not knowing when to stop with the “m” humps. (And one question mark is definitely a mismatched font while the second seems too dark to match the rest.) Bing/DE3 is much worse on the text, but much better with possums (I like good ol American possums, none of that foreign marsupial nonesense.)
The more I mess with Ideology the more I’m convinced that text is about the only thing it’s good at. I’m getting muddled messes most every other time. It does do the text nicely, though I’m not getting very crisp images otherwise. It does run a “revised prompt” - I’m not sure if you can disable that or if that is even the issue. I digress.
