# [LTX 2.5] General AI Filmmaking - How can I get Spatial Awareness for my filming locations ? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/69dab6c134953e5c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-11-03 15:29 ingested 2026-11-04 02:39

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xmvz9t/ltx_25_general_ai_filmmaking_how_can_i_get/)

## retained text



```
Hey. I'm working on something and so far the biggest challenge for me (and most time consuming) is spatial awareness and continuity. It's very difficult to generate references from multiple angles of your "sets" (locations) due to the image models struggling with continuity but even once you've done that, the moment the camera needs to move around you start getting issues with the video generator making stuff up basically because it has no idea what the rest of the room / scene is supposed to look like, even when you're using helping frames.

Let's say you have a character making a vlog and they're currently filming themselves in their room, which is supposed to have a very specific layout. Even if you give these models a start, middle and end frame they struggle to follow them because often those references are not actually spatially correct either (so they end up making a fade to transition instead of a continuous sot). Plus, they seem to invent stuff along the way. As in if your character rotates the camera quick, even if you have a middle frame the generator has enough time to "make stuff up" as the camera moves which is not supposed to be there, such as adding other objects in the room, adding a door, window which is not supposed to be, etc.

These are issues you can sort of get around but getting around this stuff consumes so much time and bad generations that you start wondering if there's a better way to get this stuff right from the get go each time.

If my character is walking on a street I want that street to look the same when I need them to walk back home. Same with their home, rooms, etc. I basically want full spatial continuity so that I don't have to worry about this. Why ? Because it helps alot with immersion.

Another thing I'd like to have is an easier way to generate references of the same location from different angels. Both Chat Gpt Image and Google Flow (Nano Banana 2) are really struggling with getting photos of the same place from multiple angles, even when they are fed references, it can take many tries until you get one good frame, and sometimes even that good frame turns out it's not from the correct location in the scene so what happens is LTX tries to make a cut, just because the camera position in that reference is like 0.5m further away than the camera position in the previous frame or stuff like that.

Is there a tool or a way (besides modeling each set in 3D which is ridiculous, might as well do the whole thing in 3D then) to get spatial awareness for all your sets (shooting locations) so that you only have to set things up once and then it will know exactly what is in a room or in a place, where it's placed and so on ? This way I wouldn't have to be worried about this whenever I need to move the camera around inside a take.

**Long story short:** Whenever the camera moves around in a scene LTX ends up "making stuff up" of what is in the scene, even when using middle & last frames because it does it in-between those frames. Sometimes it's small things which aren't suppose to be there, sometimes it's bigger things. I want to know if t here's a way to solve this issue without having to provide full 3D modeled sets of each location or something ridiculous like that.
```



document fetched 2026-11-04 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xmvz9t/ltx_25_general_ai_filmmaking_how_can_i_get
