Fragile Beauty
is my entry in narrative category for the Project Odyssey 2 AI Filmmaking Contest.
Fragile Beauty is a slow paced, thoughtful narrative short film that plays in a distant future.
The film is about society and beauty.
The idea, thoughts & background story
I asked myself, what if you are born into a world where the ordinary is obsolete?
What if humanity alone is no longer enought?
In a fast paced world where everything revolves around progress and self-optimization, where people trade their humanity and their soul for perfection, steel and cybernetic enhancements trying to keep up.
What would be left of oneself and why aren't we already enough as we are?
What happens when we can't keep up the chase, by the trends imposed on us by society?
I wanted to show that humanity isn't about beeing perfect, that it is the imperfection that defines us, that makes us unique, that makes everyone truely beautiful! - and that you don't have to be perfect in the eyes of society.
Because, with every upgrade the humans made in the film, they surrendered another piece of themselves, another fragment of their humanity.
The beauty that you are searching for isn't in what you could be, but in what you already are!
What you are right now is more than enough!
The Style
I asked myself if I wanted to go Sci-Fi Anime or Sci-Fi realistic and I decided to go as realistic as the tools I was using allowed me to.
The reasons for that were the following:
to resonate deeply with the audience, portraying raw emotions through facial expressions and body language
realistic film may reach a wider audience
to make it easier for viewers to digest and relate to, as they might see the issues reflected in the film mirrored in their own world
cultural neutrality
LoRA selection process:
Figuring out the overall style and testing a lot of LoRA's was a huge step in the process.
I tested 30+ LoRA's, a lot of combinations of them, with different strenght to come up with a unique aesthetic.
I also made a couple of custom Sci-Fi LoRAs that I heavily used throughout the whole film as well.
I came up with 3 Setups that I figured out via testing in order to get the look and the consistency around the shots.
Setup 1:
For the Androids/Cybernetic enhanced Humans and the inside environments I always used this as a base with a strenght of 1:
For the inpainting part I mixed Sci-Fi Cyborg Portraits with Skynet with a strenght of around 0.2 to 0.6:
Setup2:
For the world and the skyscraper skylines I used Sci-Fi Cyborg Portraits LoRA in combination with KFT's HLA Image Enhancer LoRA, I then inpainted the shots using:
to alter the skyscraper's into a unique style.
Setup 3:
For the Humans that are not or only slightly enhanced or have their enhancements failed or broken I always used these LoRA's:
Additional LoRA's that I used for one up to four shots were:
Hands holding Object / Hands POV and KFT's HLA Image Enhancer
Inpainting:
Inpainting was always done with the following settings/workflow
Resize mode: just resize
Mask blue: 4
Mask mode: Inpainted masked
Masked content: original
Inpaint area: Only masked
Only masked padding, pixels: 44
Sampling and scheduler: Euler/Simple for the cybernatical enhancements and every other thing.
Sampling and scheduler: DEIS/BETA for skin texture
Distilled CFG Scale: 3,5
Resize to: 1024x1024
Denoising strenght: always switched between values of 0,3 up to 0,81
I also changed the prompt to the parts I wanted to inpaint.
E.g.: I went from 'white reflective ceramic cyborg suit' to 'rusty broken....' and then inpainted those parts.
For the parts I wanted to stay the same but add more detail, I left the original prompt and inpainted small areas with denoise of 0,41.
Editing:
The hardest part for me!
As well as I tend to know FLUX and how to make LoRA's for it, the cutting part was totally new for me and super challenging and I struggled a lot in the beginning, I did cut one video before but I figured it was 12 years ago. So I had to learn all the basics again and also Davinci Resolve to bring all the parts together.
Tools and Software I used to create the whole piece were:
Civitai for custom LoRA training
Flux1.D and ForgeUI for image generation
Kling AI for image to video
Freepik for image upscaling and also for animation since they offer Kling AI animations within their AI Suite.
Elevenlabs for the narration voice and for the sound effects
Suno for the instrumental song
Davinci Resolve for cutting the film
Final toughts:
I had a real blast creating this piece for the Project Odyssey AI Filmmaking Contest !
I learnt so much and pushed thru highs and lows during the process.
The whole project took around 2 weeks time to finish and I am so thankful for all the support I got and received throughout.
I hope you like the film tho and I hope you liked the insights I shared here as well.
If you have questions about anything, feel free to ask and I will be glad to openly answer all the questions.
Much love as always!
ChronoKnight