BogDog ventures into the bottomless swamp of AI generated pics and video

[JSample]

Nicely done!

[BogDog]

A simple quicksand themed one with a woman lost in a swamp.

slip.fall.qs.wmv

[dlodoski]

A simple quicksand themed one with a woman lost in a swamp.

Dang! You had me already when she was just walking. Hehe

Lot's of variables at play of course. But the standouts for me was the wardrobe and the variations on attention to different parts of her body.

The eye contact pretty much capped everything off. Nice.

[JSample]

A simple quicksand themed one with a woman lost in a swamp.

slip.fall.qs.wmv

Once she started sinking there was no stopping her!

[BogDog]

Thank you. I learned a lot about post-shoot editing making my game-based videos for my channel and just watching movies to learn such techniques.

Now I just need to win the lottery so I can keep this up.

This is where I go: https://www.seaart.ai/ The "SeaArt Film Video Pro" model gives me the best results by far but also cost the most points to use.

The process can be fascinating. For example to get the walking away shot took several renders just to get shots of her walking. Truly it was a surprise because I did not instruct the AI to do that. There is still a lot of randomness in the creations. I took advantage and added the shot to the video.

Here is the full clip before editing. First I had to use AI to put her intact clothes back on since the walk scenes come before her fall. The original base image I began with is on the right.

ai.mod.png

Then I used the new image on the left as my base and wrote a prompt for her to stand and start walking. A lot of results get tossed for one reason or another, like the leaf-less trees suddenly are green with leaves or the model makes odd moves right out of the blue. so I have to discard that render. Another had her suddenly do a face-plant in the water as soon as she stood up for who knows what reason. Sometimes I try and try but can't make the shot do what I want and have to discard everything I rendered. That happens too often unfortunately and make this more expensive to do. Much still to learn.

And here is the full unedited walk away clip before trimming the beginning and just using the walking part: https://i.imgur.com/0ZeyGf6.mp4


Fascinating hobby. Very addictive too I confess. Hope prices drop drastically someday as competition takes effect.

[MadMax359]

here's an example of suspending disbelief--- we don't see how her blouse got open... but for me, i Don't Care!

[BogDog]

OK. Did some quick and dirty experiments with SORA 2 this evening and the results are intriguing. Definitely has possibilities and is more than 50% cheaper.

I made five 5-second clips joined together. There was some minor issue with color drift and aspect ratios between clips but I was able to work around the latter at least. Her faces changes a bit but because there is no clear shot of her face in the screen capture original image. Like a said, a quick and dirty experiment, but shows big promise.



SO here are the results, the movie scene most of us fantasized about... The Grippen Mire scene, specifically the 1959 (colorized) movie when Cecile darts and leaps through the marsh obviously familiar where the safe path is. I began the AI work with this screen capture and went from there...I also began the full completed clip with a scene from the actual movie to get the whole story started.


We all wished she would stumble in, right? Your wish is granted....sorta. Hope you find some enjoyment in it.

Grippen.mire.ai_bogdog.wmv

[MadMax359]

big promise indeed!!

[JSample]

Very nicely done! That scene was way too short in the movie.

[BogDog]

An experiment I started using SeaArt Film Video Pro v1.0 and completed with Sora 2.0. This is my first A.I. video released here using an original character.