VoyzAI

AI pipelines breakdowns
Cinematic shots at home

[VoyzAI @voyzai]

https://t.co/ZBQhKCUquD

[VoyzAI @voyzai]

@YourAlphaMom @makeugc the realism here is actually unsettling.

[VoyzAI @voyzai]

@itsSaira_1 this is the kind of AI video where the story actually carries it.

[VoyzAI @voyzai]

@TheRogueA1 this looks surprisingly clean for H3

[VoyzAI @voyzai]

@AIwithAliya pretty strong combo no subscription

[VoyzAI @voyzai]

@ZentrixHQ Thanks you too, friend

[VoyzAI @voyzai]

Two models received the exact same prompt. The result surprised even those who work with these models every day.

GPT-6 Astra on top, Fable 5.1 on the bottom. One person goes through a giant mouth in the ocean, a forest, water, a desert, an endless corridor of door frames, and shattered glass in fifteen seconds.

How to make such a comparison yourself:

PROMPT - write one scenario with a clear sequence of scenes and a single character, without variations or alternatives
LOCK - lock the character through a separate reference using Character Lock so that they don’t change between worlds
RUN BOTH - run the identical text through both models, without changing a single word
COMPARE - compare not the composition, but the details: lighting, particles in the water, the behavior of the shards, the density of the fog
PICK - choose a model for a specific task, not the "best" one overall - each has its own strengths

Which model impressed you the most?

[VoyzAI @voyzai]

https://t.co/ZBQhKCUquD

[VoyzAI @voyzai]

@YourAlphaMom yeah Astra seems to handle the underwater stuff way better.

[VoyzAI @voyzai]

@YourAlphaMom yeah that’s the part i find most interesting
whether it holds up after multiple trie

[VoyzAI @voyzai]

@ivanka_humeniuk this is getting really close to being usable for actual production.

[VoyzAI @voyzai]

Two models received the exact same prompt. The result surprised even those who work with these models every day.

GPT-6 Astra on top, Fable 5.1 on the bottom. One person goes through a giant mouth in the ocean, a forest, water, a desert, an endless corridor of door frames, and shattered glass in fifteen seconds.

How to make such a comparison yourself:

PROMPT - write one scenario with a clear sequence of scenes and a single character, without variations or alternatives
LOCK - lock the character through a separate reference using Character Lock so that they don’t change between worlds
RUN BOTH - run the identical text through both models, without changing a single word
COMPARE - compare not the composition, but the details: lighting, particles in the water, the behavior of the shards, the density of the fog
PICK - choose a model for a specific task, not the "best" one overall - each has its own strengths

Which model impressed you the most?

[VoyzAI @voyzai]

@ZentrixHQ suddenly every unnecessary follow-up question matters.

[VoyzAI @voyzai]

A guy in a bathrobe discovered he can shoot things with his fingers. Things went downhill from there.

First he decided to take out a watermelon. Then came the cans in the yard - the neighbors clearly weren't thrilled, and they're the ones who called the cops.

How to pull this off:

IDEA - pick one absurd ability and keep it consistent through the entire clip
GESTURE - choose one recognizable gesture that repeats in every scene as the character's language
SHOOT - film ordinary scenes at home and in the yard, no props, no costumes
REFERENCE - build references for the aftermath in Nano Banana - the punctured watermelon, the scattered cans
GENERATE - TapNow or Seedance add the effects on top of the real footage, preserving the movement and the gesture
ESCALATE - build the edit so it ramps up: one object, then several, then a person, then consequences

I break down how to make cinematic videos at home in detail in my guide.

[VoyzAI @voyzai]

@saurav_paswan2 this is so stupidly satisfying 😂.

[VoyzAI @voyzai]

@ZentrixHQ the fact that the fly is actually making the decisions is wild.

[VoyzAI @voyzai]

A guy in a bathrobe discovered he can shoot things with his fingers. Things went downhill from there.

First he decided to take out a watermelon. Then came the cans in the yard - the neighbors clearly weren't thrilled, and they're the ones who called the cops.

How to pull this off:

IDEA - pick one absurd ability and keep it consistent through the entire clip
GESTURE - choose one recognizable gesture that repeats in every scene as the character's language
SHOOT - film ordinary scenes at home and in the yard, no props, no costumes
REFERENCE - build references for the aftermath in Nano Banana - the punctured watermelon, the scattered cans
GENERATE - TapNow or Seedance add the effects on top of the real footage, preserving the movement and the gesture
ESCALATE - build the edit so it ramps up: one object, then several, then a person, then consequences

I break down how to make cinematic videos at home in detail in my guide.

[VoyzAI @voyzai]

https://t.co/ZBQhKCUquD

[VoyzAI @voyzai]

@BobbyBuilds_ai @BennettWaisbren I get what you mean.

[VoyzAI @voyzai]

I think she misunderstood what "finger food" means

She ate the bug

then went straight for the hand

And honestly this is a perfect example of why AI video gets weird when multiple objects interact closely.

The model has to keep track of the creature, the insect, the hand, the mouth and their exact positions at the same time.

That’s where things start getting cursed.

If you want cleaner results:

- keep objects clearly separated in the reference
- use strong multi-subject tracking
- make the interaction simple
- always check hands, mouths and contact points

Give AI too much freedom and suddenly you’re the snack

[VoyzAI @voyzai]

https://t.co/ZBQhKCUquD

[VoyzAI @voyzai]

@ZentrixHQ a fly brain controlling a car better than me is humbling lol.

[VoyzAI @voyzai]

@YourAlphaMom the boyfriend of the year 😂

[VoyzAI @voyzai]

@DylanIsaaccwgn @BennettWaisbren glad I’m not the only one.