Claudia's little secret in 2026. What can we achieve with AI ?(WARNING AI!)

[Grumpf]

I wanted to share that with you guys:

I did learn how to prompt for stable diffusion renders, everyone can do a 5 second shot of naked elf dancing in the woods, the challenge here was more to see how to keep different characters consistent through minutes of render. There's no real solution actually. So, of course, I had to try... First video below you can see an example of easy re-render of a 20 YO game. Second video is actually where I suffered a lot as I tried to keep it consistent. Won't bore you to death with the details but yeah... Anyway, it's not a game or a mod but I though you guys may like to see this and maybe, for some, change your mind about AI slop.

Cheers !
Grumpf.

 

 

My story though, will be something different, you can grasp an idea with the credits here: (check links below for HQ)
 

 

 

My boring explanation below, can be skipped  

 

 

Today, I have finished and published a 3½-minute AI-assisted fantasy film sequence following Païka, a fallen goddess, and Anna, her human companion, through friendship, adventure and mortality.

This is the opening and end credits sequence of a much larger story I've been developing. It includes original characters, animation, music, lyrics and subtitles.

Watch the film on YouTube

Download the high-quality 3K version on Internet Archive

 

Also CH1, more or less not a draft anymore. 5.1 Surround if you can !
https://archive.org/details/1024-audio

 

THere's a PG13 version of it but somehow, I think you don't really care about it !
https://archive.org/details/ch-1-pg-13

 

Now the really boring stuff, you were warned.

The journey

When I started this project, I knew almost nothing about local AI generation. I couldn't install PyTorch. I didn't know ComfyUI. WAN wasn't part of my vocabulary.

Then, one obstacle at a time, I learned what I needed because the next shot required it.

The final pipeline eventually involved ComfyUI, WAN, DALL·E, Qwen, Photoshop, Flowframes/RIFE, FFmpeg and Subtitle Edit. Not because I set out to learn eight different tools, but because every stage of production presented a different problem.

Generate a consistent character. Animate her. Repair a face. Preserve detail. Interpolate motion. Upscale. Assemble nearly 10,000 PNG frames into a lossless master. Encode a playable version. Add subtitles.

And, occasionally, convince an AI that a medieval statue should have exactly two legs.

What I learned

The biggest lesson was that prompting is only a small part of making a coherent film.

A beautiful still image is not necessarily a good starting frame for animation. A face that looks slightly wrong at low resolution will not magically improve after upscaling—it will usually become spectacularly more wrong. Objects teleport. Glasses multiply. Trays ignore gravity. Squirrels occasionally float.

The workflow that finally worked was roughly:

WAN at 8 or 16 FPS → ComfyUI upscale to 3072×1728 → Flowframes/RIFE interpolation to 64 FPS → lossless FFV1 master → HEVC delivery encode.

Every scene was generated and interpolated separately before final assembly.

 

AI-assisted, not AI-generated

I deliberately call this an AI-assisted film.

The images and animation were generated with AI tools, but the story, characters, direction, shot selection, editing, pacing, rejected generations, workflow design, repairs, lyrics and final decisions required months of human work.

AI generated possibilities. I decided which ones belonged in the film.

There were countless failed shots and discarded versions behind these three and a half minutes. The final result is not perfect, but it is finished—and it is the film I wanted to make.

Seven months ago, I couldn't install PyTorch.

Now I have a film.

That's a pretty good upgrade.

 

[belegost]

So this is the reason why the price of water and electricity in my constituency had tripled in the past 6 months.

Edited July 29 by belegost

[GlitchyPink]

Impressive, but it's giving me a bit of a headache (as if I've been staring cross-eyed for too long).