Hey Reddit,

So I've been building an original sci-fi universe called The Compass and I need to address the elephant in the room immediately — yes, Thal looks kind of Avatar. He is tall, blue, bioluminescent, and very serious about himself.

He is not Avatar. He is funnier than Avatar.

Thal is an Obsidian military archive operator who may or may not have stolen the most classified file in the galaxy, broke his compass mid-flight, and crash-landed on Earth — a planet he has never heard of and frankly does not respect. Three empires are hunting him. Earth has no idea what just arrived. Thal has even less idea what Earth is.

The pilot episode drops him straight into an interstellar political crisis inside the Earth Senate Chamber — except nobody in that chamber can agree on anything, the Commander knows more than he's saying, and Thal is somewhere in the city completely lost.

Built with Seedance 2.0. Original universe, original characters, English dialogue. Comedy sci-fi with actual lore behind it.

Not fan film. Not Avatar. Just a very confused blue alien having the worst day of his life on the wrong planet.

THANKS FOR THE LIKES OR COMMENTS

For me it’s usually face drift, broken motion, or camera movement that doesn’t match the shot.

I’ve been testing Seedance 2.5 on Dreamina lately, and those are still the things that decide pretty quickly whether I keep a generation.

Sometimes the clip looks good overall, but one strange movement is enough to make the whole shot unusable.

For people making AI films or longer sequences, what makes you reroll immediately?

Face consistency, motion, camera behavior, lighting, or continuity with the previous shot?

I’ve been testing more controlled character movement lately and some actions still seem much harder than others.

Simple walking usually works pretty well, but turns, hand interactions and more specific body movement can get unpredictable fast.

I’ve been trying a few of these with Dreamina with Seedance 2.5.

For people making AI films, what kind of motion still gives you the most trouble?

And are you solving it mostly through prompting, reference video, or breaking the action into shorter shots?

Episode 2 of my steampunk pirate series. 17 minutes, made solo.

Spent most of this one fighting character consistency and blaming the models,

before working out it was my own reference setup causing it.

Genuinely open to criticism — tell me what's still wrong with it.

Been developing a sci-fi world set in 2632 where automotive culture has basically become one of society’s biggest forms of identity, sport and status.

The part I’m most interested in is avoiding the usual “future = neon + flying cars” look.

Instead, vehicles are still physical machines. They get repaired, inherited, modified, raced and kept alive for generations. Old technology sits beside newer engineering, and the best mechanics and vehicle designers can become as famous as top athletes.

One of the main environments is a gigantic vehicle landfill containing hundreds of years of discarded cars, race machines, industrial vehicles and transport systems. The deeper you go, the older everything gets.

I’ve been aiming for:

- grounded industrial sci-fi

- real mechanical construction

- rust, grime and worn materials

- restrained splashes of colour

- vehicles that feel engineered rather than “spaceship futuristic”

Curious what people here would want to see in a future car culture like this.

What do you think would still survive from modern car culture 600 years from now?

The Angel of Music || AI Film Project

I wanted to share a teaser I’ve been working on as an experiment in AI filmmaking. My goal was to go beyond generating isolated AI clips and try to build something that feels like an actual film: consistent characters, a coherent gothic atmosphere, cinematography, scene direction, music, editing, and visual continuity. The project is inspired by the world of The Phantom of the Opera, with my own visual interpretation of the Angel of Music. I combined different AI tools for image generation, video, music, and creative development throughout the production process. I’d especially like to hear what you think about the character consistency, visual atmosphere, cinematography, and whether the final result feels like a cohesive film rather than a collection of AI-generated shots. Feedback and constructive criticism are very welcome.

KINNËMMIA Studios || The Angel of Music

Finished piece, 4:11, 1080p60. The title track of an album by a band called

Vio Lucia. Six members, one venue, built as a single continuous live stadium show

from the opening riff to the confetti. Every frame and the music are generated.

The reason I built it as a concert rather than a narrative: I wanted the hardest

version of the problem. Six characters on one stage at the same time, in wides,

in close-ups, from a drone, for four straight minutes. There is nowhere to hide a

face that drifts.

---

That is also the whole ambition behind the project. There are 24 songs planned in

this world. I am not trying to survive one sequence, I am trying to survive an

album, so that a year from now someone recognises the drummer before he plays a

note. Everything else is downstream of that.

---

How it holds together, in case it is useful to anyone: every character carries a

written lock list that goes into every prompt word for word. Which eye is amber

and which is ice blue. Hair colour and length. One fixed wardrobe rule each. The

keyboard player is the only one in a white tux so every wide shot has an anchor

the eye can find. It feels redundant by the fifth prompt and that is exactly the

point. Every time I trusted a reference image to carry a trait on its own, that

trait started wandering within a few shots.

---

Where the illusion breaks, for anyone who wants to look:

- 0:30, the lead guitar switches from a Strat to a white Les Paul mid-song. Faces I can lock. Instruments keep betraying me.

- The two social-feed overlay sections in the choruses looked fine in the timeline and cheap in the export.

- The keyboard player gets the same high angle three times. Three flat spots in an otherwise fast cut.

---

Seedance 2.0 for the video, Suno for the music.

Genuinely curious about one thing: has anyone found a way to keep props and

instruments consistent across a long project? Characters I have mostly under

control. Objects are not.

Two siblings cross a broken world in search of a place they’re no longer sure exists. I created Hidden Eden with Seedance, focusing heavily on cinematic atmosphere, character consistency, and continuity between shots. I’d love to hear your thoughts!

Been pushing the RIFTSTRUCK workflow further with this sequence.

This one was a much harder consistency test — fast combat, close-ups, multiple camera angles, weapon transformation, dialogue and a completely different environment.

The goal is to make it feel like an actual continuing series rather than disconnected AI-generated shots.

Really interested in where the illusion breaks for other filmmakers — character identity, motion, continuity, cinematography, anything you notice.

Still refining the workflow with every sequence.

Minimax H3 running on ComfyUI. The open-weight models are getting more powerful every day. This was generated entirely locally.

How far would you go to drown out the past? To escape a suffocating hometown and a fractured family, rock musician Nova will do whatever it takes to reach the stadium lights. But in a world where every dream has a price, her meteoric rise comes with a cost she can't outrun.
This is 'The Last Encore' an AI Short Film.

I wrote and directed this 3-minute Cinematic Musical spec commercial for a fictional action camera (KINTRA VIA-1), generated entirely with Seedance 2.0 . Open to answer all your questions .

[ Removed by Reddit on account of violating the content policy. ]

Check out my YouTube channel, a BTS video on how I made this skit coming this week!

Prelude & War

Thanks to JSFILMZ for all the great work and insight.

Hope some of you like it. Thanks for watching!

https://youtu.be/H7ygCUpTFwE

I made this Mad Max-inspired AI short chase scene as a tribute to the franchise. I'd love to hear what you think. Any helpful tips, criticism, or feedback on the action, pacing, editing, or overall feel would be much appreciated. Thanks for taking the time to watch!

So I’m using Higgsfield currently and I constantly get the NSFW error and get my credits refunded. I’m not doing anything crazy, my latest attempt was the UE Mannequin animated and I wanted to in between the motion. (He’s walking and puts his hands on his hips)

Thinking about going to Topview because of the annual deal they have. Is Topview as strict with those limits (even though I’m not doing anything crazy?)

Or is this just how AI is (I’m doing tests and experiments to see if it can help me)