AI filmmaking isn’t always predictable—but that’s part of the magic. In this lesson, you’ll learn how to handle unexpected results in the Movie Machine, troubleshoot common issues like inconsistent visuals or voices, and adapt your story when AI behaves differently than expected. You’ll also discover practical fixes, from refining prompts and regenerating frames to adjusting director’s notes and using reference images. By understanding AI’s quirks and limitations, you’ll gain the confidence to turn surprises into creative breakthroughs.

Below is the transcript of the session with chapters. Please note that the transcription was done by AI so some inconsistencies may occur.

Introduction to AI's Unpredictability

Dealing with unexpected moving machine behavior. AI isn't always predictable. It's probabilistic, which means if you give it the same input twice, sometimes it has a different output because it's not deterministic in nature. It's probabilistic thinking, oh, it's probably this outcome it wants each time.

Right? And it's a probability distribution rather than, it being a defined outcome. Sometimes this means that AI results don't match your intent. Maybe the character doesn't look how you envisioned a shot generated incorrectly, or sometimes a shot is just rejected outright. Luckily, you don't have to pay if a shot's rejected and isn't generated.

But these things can happen, and part of working with AI is exploring and adjusting your story so that it can get around some of the issues that AI can have at times.

Common Issues with AI Results

So ultimately, results don't match your intent when there's missing or unclear character seen descriptions. This is the most common issue we have.

AI interprets details differently, so the details are there, but they just need to be mentioned differently. There's complex shots with too many elements, so it confuses the ai, or there's some system or model limitations, which we'll cover those. Ultimately, we're here to support you.

Support and Troubleshooting

I'm not expecting you to know how to troubleshoot all this, although I'm gonna give you the playbook of how to yourself.

But reach out to us at hello creator tv. What's really helpful is to share these four things, what you wanted, what your intended outcome was. So what were you trying to do? What did a character, you wanna have blue hair, but it has red hair. Cool. What happened instead? Okay, so again, got the example.

You wanted to have blue hair. Instead, this character has red hair. So then what did you do? Like we understand it has red hair. You, we understand you wanted do blue hair, but what have you tried? So did you click edit, which I'll show you to do, and if you edited that character to try and generate a new version of it, what did you type in?

That would be very helpful for us to know so that we can troubleshoot with you. Then screenshots or short screen recording is great just so that we can get in or even just a link directly to the shot and scene you're struggling with so that we can dive in with you and not have as much back and forth so that we can ultimately resolve your issues faster.

Ultimately we've got your back. If there's a major issue. Cause with moving machine we refund new coins. We're in testing right now, we're learning. We don't have an official refund policy because we incur a lot of AI costs. If you just generate a hundred minutes of film that you're unhappy with, we'll be like, come on, a hundred minutes of film and you wanna refund the whole thing.

Why did you keep going? But if, yeah, you generate a few shots and you're like, this is way off, we're very happy to look into it. Very happy to make amends. AI is always gonna do some monkey things, but we're very happy to help.

Current Limitations of AI

So there's some current limitations to be aware of. This is, as of right now, when I'm recording in November of 2025, camera angles, there's certain angles or complex shot perspectives that just may be difficult or inconsistent.

We're working on this, but for right now, shot angles can be somewhat challenging at times. Styles, cartoon animation is like amazing. Photorealistic is not fully there yet. It's pretty good, but there's just, it feels a little ai. It's not photorealistic, it feels a little AI photorealistic and we're working on changing that.

So it's like really feels authentically Photorealistic voices. Young children or baby voices may not match expectations. This is because of a lack of training data which makes sense. If you do have a children or baby voice you want in your film, I recommend uploading that. When it says modify voice on cradle wood, you can actually upload.

Your own voice, which is what you gotta do. So you could upload a voice of a baby or a young child, and I think that will get you much closer to what you want. Versus if you just tell the eye to make a baby voice, it's gonna do a bad job. Also, sexual or extreme violence can cause generation issues. And we'll talk through how to get, how to maybe get around these and how these things work.

But ultimately, creator would. We don't really allow adult content to be monetized. So content that includes people showing genitalia or having, very much like graphic sex on screen, graphic depictions of sex, but even, putting that aside, sometimes less graphic depictions still run into issues.

Knowing these limitations help you plan a project that's seed in today's tech. You can work around these limitations science by just adjusting your story. Maybe it fades to black or is closed door. If it's a steam museum, maybe it's an art or a film style that you choose for now and then make a new film later and photorealistic.

So ultimately we maintain an updated blog with current limitations and workarounds at status Creatorwood tv, status Creatorwood tv. We'll link it here.

Unexpected AI Surprises

Unexpected also isn't always bad. Like sometimes AI can surprise you, but in a good way. I remember I was working with the creator once on a black and white film project, and we've since made it much more reliable for black and white.

It's very good. But there was like random C characters who were color in the middle of the black and white film. So it was a black and white film with oh, this character is colorful. And it was actually pretty cool and would normally be very hard to pull off without ai. But with AI it was like easy and really cool.

So it was like, oh, I didn't want this, but I like this. So that's really the fun part. Okay.

Self-Troubleshooting Tips

So we are always very happy for you to contact our support team, but it's faster if you are able to quickly. Solve your own problem. And that's what we're here to teach you how to do. So before contacting us, 'cause we're gonna be looking through these things too, and basically trying to figure out the same things with you.

Are your video prompts clear, specific and aligned with what you want? We show you your video prompts, the storyboard step, and sometimes the video prompts just aren't exactly what you want 'em to be. We might have a wrong camera movement, might be describing a character incorrectly, and most of the time, again, it works really well.

But if you're seeing issues. That's a good place to look. Next is the character descriptions. Are they detailed and accurate to your vision? Sometimes specific things about a character ma matter, like where a scar is placed. Being able to not just show that in the image, but also say that in the character description will lead the AI to following your instructions with much higher fidelity.

And I'll show you exactly how to do this soon. Now, if everything checks out, please reach out to us. We really wanna help. So if things are looking good. At least in principle, but not coming out good in practice, we're here for you.

Advanced Troubleshooting Techniques

Some quick fixes are regenerate, start or end frame. Sometimes if your start end frames are inconsistent, you just need to give it another go.

If you tell I to do something a hundred times just like a human, it's gonna make mistakes a few times. So asking you to do it again sometimes be helpful. Now, I don't want you generating more than two or three start frames, without making any changes past a certain point, especially once you get past five start frames.

If you're still having issues, just reach out to us and we'll help. Sometimes reducing shock complexity helps if you have four characters in an image or start frame, or let's say two objects, a setting and two characters. It can be a lot for the model to take in, and you can see the start and end frames not be as consistent and the videos not be as consistent as a result.

Sometimes just a different camera angle or simplifying the prompt can give you what you want. Or switching to a different AI model for that shot, which we actually do automatically for you. But in the future, if we like, allow you to select your model selecting different models or just selecting different options, like we'll have maybe high quality or lower quality or some, something of that.

Sometimes different routes can get you different options. So we do this a lot in the backend for you automatically with the switching different data models for character and style notes. Adjust those director's notes, specifically the physical description of the character to be more clear. You could do the same thing for settings and objects too.

You can also just create a new character version rather than adding the original. Sometimes, having the wardrobe be a different version or, hey, maybe you want an older, any younger version of the same character. These things can work. And you can also use reference images for more control. You can download any image from Creatorwood and make it a reference image for another one.

So maybe you have two twins. You wanna make one twin, the reference image for the other twin. You could also include images from outside of Creatorwood pictures of yourself, pictures of friends that you have, their permission. All of that for audio issues, regenerating audio lines cost one coin per line.

I would test different versions and voice styles. You can actually add emphasis into the actual dialogue. So if you wanna have someone say something loudly, make it all caps, like all of it caps like hello all caps, and you might see the AI emphasize that and actually say it louder. And also what we're adding is the ability for you to voice act.

Where in the end, once you've generated your dialogue, if you don't like the intonation, you can then say I would like you to talk this sentence like this, right? And then you could say that and the AI will mimic your imagination. Pretty cool.

Preparing Your Story for AI

With all that said, let's move on to our next session all about how to prepare your story for the move machine.