The new versions are also available at https://ideogram.ai. Ten free prompts per day.
(Like the poster.)
Dall-E 3 tends to give good results of mixed creations using the word “hybrid”. Seeing a similar image posted, I tried for Imperial Star Destroyer cake/pie hybrids, and it “got” what I was looking for. (I then tried for bowls full of Death Stars).
(Asking for “bowels” full of Death Stars still gets you bowls. Asking for x-rays of bowels full of Death Stars, however…)
So I tried creating a version of a Tsuchigumo spider from Japanese mythology (specifically when a samurai split one open and human skulls spilled out) in Bing. It is sometimes depicted with horns, so I was going with a cow spider hybrid. But any descriptions of violence or gore were either blocked or just not understood. Using the word “skull” were blocked or “cute filtered” but calavera is allowed. Never got anything like what I originally planned, but ended up with lots of images in various styles of cow-spiders barfing skulls.
(This is what a Tsuchigumo toy I have looks like, for reference.)
My favorite image turned out to be a cow doing a spit take about a weird skull-spider on its back.
More images in various styles.
As I mentioned in a different thread, Night Cafe has added Wan 2.1 video to their Pro options, along with Kling 1.6, which I hadn’t heard of before.
Kling has some interesting features, such as the ability to upload up to different photos and use elements from them and have the subjects interact.
I signed in for free trial credits and dug around in my archives for ideas. I picked two of my actual photographs and one AI render
And asked for a kitten riding a monkey in a desert.
The kitten is fairly similar to the photo, the monkey not so much
Kling AI’s graphics are pretty good, but I must say, the message is spot on!
Was the original image made at Kling, or imported?
Cross-posted from a different thread, Luma Dream Machine now has an option to “dream” audio for the video clips that you create. I had it make audio tracks for all of my previous image-to-video tests there, here is a reel of a few dozen of them.
It seems like it is using the “feel” of several different languages in creating its gibberish audio, not just English. (I’m assuming it is as much gibberish in all languages as it is in English.) It allows you to describe what kind of audio you want, but I let it decide everything on its own in most of these
Kling
I just noticed something: when you click on the download video link, it gives you the option of downloading a watermarked version or paying for a clean version. It then just opens the video in a new window and you have to right-click or long press and choose save to device. But you can do the same right-click or long press on the “preview” video and download it the same way, in full resolution without the watermark!
Five still images created in Bing/Dall-E 3, and animated into video clips with Kling AI 1.6.
The new image generator inside ChatGPT-4o, reportedly a Dall-E variant, handles text a LOT better. And it’s also a great improvement on creating consistent characters, not 100% there yet. I tried this. It took a bit more work than just feeding it the original xkcd comic and a prompt.
From xkcd 1024:
Duty Calls.
(Made with plain old Bing.)
Summary
photorealistic 3d stick figure sitting at a computer typing. it has a speech bubble that says “someone is wrong on the internet”. grainy disposable camera snapshot with shallow dof and forced perspective.
Here’s that same prompt with the new “Juggernaut Flux Lightning” model at Night Cafe:
from xkcd 1553:
I’m trying to streamline the process of selecting random xkcd comics to translate using as few prompts as possible. I don’t think it actually used the python interpreter like I asked it to this time.
Indeed, the ChatGPT image generator looks much improved—accurate text and knows what a Korat cat looks like:
xkcd 652. Took a lot of retries on this one, this was the best.
I had to tweak the thought bubble text on this one, but I’m really liking how ChatGPT 4o/DALL·E interprets my concepts graphically—and it’s a dead ringer for my cat, Benny! I usually lean on Midjourney for its richer, artistic style, but DALL·E tends to nail the accuracy of my prompts more consistently.
I see that it still has the problem other AIs have understanding a snake. You tend to get closed loops like that.
You just have to imagine that the distal third of the snake’s body took a 90 degree turn down into a hole in the ground. 
Sontarans and Raxacoricofallapatorians can crossbreed?