AI image generation is getting crazy good

Totally fake looking. His mouth is not open and he’s not angry. :wink:

On a more serious note, that is pretty impressive rendering. Well done! And I bet that if idjit trump ever saw those, he might well decide he wants Conway Twitty hair for real.

Some fun with Gemini video.

Today’s Snuffy Smith, a direct conversion.

Make a live-action video based on these storyboard drawings.

A frame from today’s Mary Worth, a direct conversion.

Make a live-action video based on this drawing.

Bit of a glitch there. I tried again with two steps.

Convert this to a realistic photo with no text

Fed the photo into a new chat:

Make a video. They angrly smush ice cream on each other’s faces.

More playing with comics as storyboards. It grasps the concepts of some better than others. Some have the order mangled. Others delete some of the dialogue. But still pretty impressive results with such little input.

My favorite

Others

(Out of the 10 free videos per week, I’ve made 13 since last night.)

Reminds me of this “tribute strip” for Pearls Before Swine, that I composed via cut and paste:

So perhaps a bit disturbing.

On the Kling web site sample page I’ve started seeing variations of output from prompts like “Jenna Ortega smiling and crank-starting an off-camera olde timey car” or “…trying to empty the last bit of ketchup from a container just below the camera frame.”

I think you probably get the idea.

Ha, this actually worked.

That’s really cool.

Kudos to Randall Munroe for allowing this sort of thing.

That’s the first time I’ve managed to convince it to allow an image after it had blocked the attempt due to invalid copyright concerns.

For the last few images I’ve done, Gemini has been downloading as jpg instead of png and (more importantly) hasn’t included the 4-pointed star watermark in the lower right.

Create a realistic portrait-orientation photo of a possum unzipping itself exposing two stacked weasels inside.

Copilot is still using their highly intrusive new watermark.

Highly-obtrusive but also extremely easily removed. Just make sure your picture has a little more background on the top than you need, and then crop it out using absolutely any other image editing program.

You take what they give you, you may get more than you need, you might not.

Can you add “include clear, empty sky at the top of the image” in the prompt?

Google actually has a really sophisticated watermarking system called SynthID that embeds the watermark in a way that’s practically invisible to the humans looking at it but is easily detected by programs even if the photo is edited, cropped, or rescaled. It embeds the watermark somehow at the pixel level in a way that’s algorithmically readable. That’s a really valuable tool in a world where AI image generation is good enough to present as real, and I think they make it available for other people to use the tech if they want.

I’m not sure how I feel about blatant “made with AI” watermarks like copilot apparently does now. I appreciate the disclosure, but it’s so distracting and ugly that it ruins the aesthetics. I think a relatively subtle one that you can see if you look for it but doesn’t jump out at you is probably the best compromise.

Google’s little semi-transparent star was a good version of this. I don’t know why they stopped using it.

Haha, does Comics Curmudgeon know about this yet? So many Mary Worth storylines call out for improvements of this sort.

Yes, right now Google’s Synth ID is used both by Google (in Gemini) and Open AI, although you can’t universally check for a Synth ID. A Gemini Synth ID can only be detected directly using Gemini (just upload or paste the photo and ask if there’s a synth ID present.) Open AI has its own synth ID detector, but it’s not through ChatGPT, it’s its own web page.

Wanted to see if Gemini could figure out what was going on in this comic (I asked for a realistic photo, didn’t explain anything.) It did pretty well, but thought the cat was going to slingshot a computer mouse.

Google Flow uses Nano Banana (Normal/Pro/Lite options) and has never included the visible watermark in all the time I’ve used it