Has anyone used ChatGPT to create images of themselves as Batman, Robin, etc.?

[njboywonder]

It's incredible for sure. However, there's something (no idea what as even other photos I create from that one (different henchmen...)) don't work as well. Making that one first had me thinking all would be as good. But starting with other, even similar, photos yields vastly different results with varying problems--not following directions, virtually no action, distortion, etc.

Will add to the YouTube page if I continue to get good results. Will try to do some based on specific scenes from specific episodes.

If anyone might want to see themselves in one, feel free to PM me.

Thanks.
Brett

[antman67]

I Like using AI, I've seen every 1966 Batman and Robin image I think that's out there and finding photos never seen before are rare so when you use AI at least the images are new.

[njboywonder]

Is the photo one you did, antman? Is that you as Batman?

[njboywonder]

And, having just looked at my YouTube post, no idea why the quality is so poor there but the actual video is perfectly clear.

[njboywonder]

Also, if anyone with any costume (or I imagine I could add the costume) wants to be integrated into a video, PM and let me know what you'd like to see.

No guarantees but fun trying.

[njboywonder]

A bit of a disappointing update...

Apparently I had some great random luck with my earliest attempts as subsequent efforts have proved mostly disappointing.


Despite extremely concise wording, issues with Chat GPT include:

--constantly indicating requests violate policies when they are benign (make a waist high image full body length, change a person's clothing, correct Robin's boots, etc.)

--going wildly off instruction, such as drastically changing ages

--missing, as I already mentioned, Robin's tights or his eyes

--ignore the requests completely--doing something completely different (changing people instead of clothing)


And with Hailuo include:

nowhere near the naturalism of the results in the link I sent

lack of photorealism

inability to deal with props (Even if I change a Chat GPT image to have one person holding a prop, it won't be used naturally. Instead, the prop appears out of nowhere and acts on its own.)

clothing elements changing during action

difficulty changing settings, going through doors.


Likely more but I deleted all of the many complete failures so don't have them as references.

If anyone is using these tools (I'm surprised more here aren't.), aside from being interested in seeing your results, I wonder if you are having similar issues and especially if you found ways to overcome them.

Thanks.
Brett

[njboywonder]

That is incredible! Just imagine where this will all be in five or 10 years. Future generations may never see real actors on film again. I can imagine once popular actors and their estates will be licensing likenesses. Perhaps there will be an all new genre of popular AI actors.

Re: your last comment...

I was immediately struck by the way AI interpreted the mens'/henchmens' expressions. I gave no instructions but in the video in the link the two men take the same delight in capturing Robin as on the show. For that reason, and others that avoid the issues listed above, while I've tried a good number of others, I've continued working with that original photo, hoping it will give me something more quintessentially series-like.

So so far, these are my favorite AI generated 'actors...'

Brett

[njboywonder]

Still nothing I love.

Some more suggestions to help others trying.

Download the Chat GPT images you like and delete all of your projects (Possibly the one in question might be enough depending upon how each project differs.). That seems to help it stop making the same mistakes and stop telling you you violated policy. I got a reasonable version of almost everything I requested that was first or even repeatedly rejected by doing this.

For Hailuo, leave out the adverbs. Use the most concise request you can, possibly using nouns with more implications to get the action, facial expressions, and overall tone you want.

Leave out notes about expressions or feelings. Writing that someone should smile can make everyone smile, etc.

It would be very interesting (and perhaps helpful) to see if someone can accomplish something I can't with the same source image.

[antman67]

AI is hit or miss? Bing seems to be decent

[njboywonder]

I didn't know Bing had one. The only other I knew of was Canva. I discovered last year, I think but it wouldn't allow licensed characters or working from my photos. Will check it out, thanks.

Are the photos you posted of Batman you/based on you?

Brett

[njboywonder]

I had a pretty great result with another one, though it's very similar to the other two.

I did something for a friend involving Under Armour.

As I noted, sometimes the program draws from other things in your work history.

Many times it will follow your directions but the results still aren't what you want. When that happens, I just ask it to try again.

This time, it randomly changed the (hench)men's clothes to Under Armour. But, it happened that the photo gave the most exacting image of me so far and so I had to try to work with it and very luckily, the video not only held to it but gave me one good compete result.

So, I couldn't not post that one as well.

The guys' Under Armour is the same so I'm convincing myself it's kinda like a 2025 version of the series' henchmens' matching outfits...

As far as the trouble with props goes, I have what would be a very cool simple idea that I'm trying to make work. I have the photo, but so far the site isn't having it, making a mess of it for no good reason every time. We'll see how much I'm willing to devote to it.

Thereafter, I'm going to move away from this series of source shots from the parking lot and try for something more series based--hopefully using one of the photos of one of my Riddler costumes.

Brett

[njboywonder]

Still having a very difficult time with both sites.

Chat GPT refuses to get images correct and I've made dozens of attempts at the same two Extremely frustrating and of course cost ineffective.

Hailuo has done incredibly intricately detailed work but of things I didn't request. Would be so much easier to create Robin spinning on a wheel than to have him fly off and maintain the necessary details of that. But that's the kind of thing it does...

I finally went for an extremely simple video from an early/existing shot, and it only took three tries (Slid up the Batpole, fell down the Batpole, finally slid down the Batpole...) and posted it.

Also, and I know nothing about YouTube, somehow my third post got like 75 views in the first minutes and a hundred plus more since then. Not sure how/why.

Finally, I just discovered you can ask Chat GPT questions re: why things aren't working and how to make them work. Starting on that now and hoping it will help.

Brett

[njboywonder]

More things I realized since I last posted...

Make the photos horizontal (IN MY CASE, IT VIRTUALLY ALWAYS DEFAULTS TO VERTICAL SO I HAVE TO INSTRUCT IT). Of course it gives a larger on-screen video image.

Anticipate the positions of the characters' best starting point in the video when deciding the photo compositions. Something as simple as taking an extra few steps or turning around wastes a lot of time when you're limited to 6 seconds.

As I Mentioned, starting with a photo that has a character holding a prop didn't yield good results with video. But, very unexpectedly, introducing a character (not present in the photo) holding the prop into the action of a video has been working well.

I gave up on doors. I now divide action. One video showing the outdoor part with the second the continuation indoors. It obviously also allows for more action since there's more time covered.

You'd think a story board like this is an obvious smart choice but it's so incredibly time consuming trying to get one video right that it's a risk to need two to tell your story.

When you create an image of Robin that randomly comes out looking like James Dean (complete with leaning), the video treats him like James Dean. He had no reaction, was too cool, for everything happening to him...

And I'm spending too much money on this... Luckily I messed up my knee and can't do what I'd normally be doing (wildlife photography) for a while so it's good timing. But still a few hundred dollars for something like a 1% success rate.

Brett

[BiffPow]

But still a few hundred dollars for something like a 1% success rate.

Wow! Had no idea it was that expensive. But, watched your videos on the YouTube channel. Enjoying seeing the results. Just hate that it’s costing you that much.

[njboywonder]

But still a few hundred dollars for something like a 1% success rate.

Wow! Had no idea it was that expensive. But, watched your videos on the YouTube channel. Enjoying seeing the results. Just hate that it’s costing you that much.

The success rate is really, really poor. It could take dozens of attempts to get the photo right and dozens for the video as well. Often the video program just never gets it right.

Were that not the case, the costs would be nominal.