————
This short was painful to watch. The AI “actors” were so creepy and the continuity details were so bad that it was hard to concentrate on the looping. 

I will say that the AI looping was very bland and generic. It did fit the scene, but didn’t add anything to the experience. There were also large gaps where there were people in the immediate background, but no sound from them at all — eerie.

I liked the fulness that the human looping added, like the unseen person interacting with the cat in the beginning, and the genuineness of the laughter of the children.

Please please please use human actors. It was an effort for me to continue watching in order to do the comparison. There was no connection at all between characters, and their faces did not reflect emotions accurately (or match their voices at times).

My dos pesos.

Karyn O’Bryant
————

I really enjoyed this! Obviously there are some shortcomings with using AI from top to bottom, but I definitely noticed the difference between the Rebeloop version and the original. The first minute, a collection of shots from different perspectives, and especially the bar scene, had more ambiance with Rebeloop involved. Would the average viewer notice? Probably not. Will your sound engineers and post production folk see it? Absolutely. There were a few moments while they were talking in the bar where the background audio felt too loud/distracting (maybe because I was listening for it) but otherwise, thumbs up. 

I think there is always room for human input on any project, and AI creators would be silly not to consider using that. But cost-wise, it doesn’t make sense. If the AI version is 66% as good but a fraction of the cost, most people would probably choose to cut that corner. Not saying that some wouldn’t, but I think projects like these will show well on social media instead as ads/experiments, especially targeting creators and filmmakers. The industry is changing so quickly, it’s hard to say how things will look, but I always like to say it’s worth throwing it at the wall and seeing what sticks!

Dylan
————

  1) I think next time give us both versions but don’t reveal which version is which. I find it really hard for me to listen or see something without bias when someone tells me something about it that I may not have normally picked up on?…

2) The AI version to me felt very ‘flat’ for a lack of other descriptive words. There didn’t seem like there was any dimension to it, and watching it almost made it seem like watching a bad video game scene to me. 

3) The version with the background noise added felt more ‘real’ to me, more rich, better fleshed out. I can’t pinpoint exactly why but I did recall one scene in the bar where there was a party of people cheers(ing) each other and AI version had no life and no context in that scene whereas the with rebeloop version felt like it was a normal scene that you’d hear in a bar. 

Overall I just felt like the AI version didn’t feel ‘real’ but I wonder if I would have thought that if I didn’t know it was the AI version…. hmmm….
Asha said she liked the rebeloop version because I guess this is what you all worked on last time so she enjoyed watching it come to life. She also said that watching the rebeloop version makes you feel like you’re in the scene and ‘alive’ versus just watching something and being disconnected from it. 

 I think for both of us it really is hard to describe but it’s just the ‘feeling’ that the non-AI version can produce. It’s subtle but it’s there…

Teresa
————

They are noticeably different, yes. In fact, before playing them, I wished you had sent them not saying which was which, to give us a blind taste test. But it’s pretty obvious: 

Specifics: the AI doesn’t match anyone visible onscreen. Lip movement, in particular, is missing. Also, I think AI overcrowded the school grounds in the first scene, it’s way too busy.

That said, I think Rebeloop overpowered the bar. The specifics (lips, laughter) were a match, but there was too much distinction in the background dialogue, to the point of being distracting. The AI succeeded at creating ambience. But that probably just needed you to give it another pass in editing.

The AI loopers themselves “sound” good. When mixed into the background, they sound like natural voices. The main characters don’t, they sound like AI. Now, you could surely train the AI to solve the specifics, but I suspect if the AI were to match specifics as well as Rebeloop, that false sound of the voices would come through and it wouldn’t work as well. (Of course, for either AI or RL to match those lip movements, it assumes that the background AI “actors” are actually mouthing words, when it’s possible AI just has their lips moving in a way that looks like speech.) Which brings me to a side point: I would recommend thinking twice about supplying these samples to an AI filmmaker, as they will likely be exploited to further train the AI.

If you’re talking to an AI content creator, you’ve already lost the argument. Yes, a human loop group creates a far superior product – but so do human writers and actors and animators. And in the case of this film, at least, adding a human loop group is, as they used to say, putting a forty-dollar saddle on a twenty-dollar horse. The entire look and sound of this film is exceedingly inauthentic and false. If the filmmaker is looking at any piece of this and saying “Yep, that’s excellent, that’s what I want to put out into the world”, then their aesthetic sensibility is a loss – any improvement by the addition of a human element will surely not be visible to them. Plus, I mean… this thing is a mechanical abomination, fixing one element won’t save it. And like I was saying, if you educate this person on what their AI missed in the looping – they’ll just train the AI to do what you said.

-Jason M
————

1. In the AI version, street scenes felt DEAD. Struggled to really hear any kind of ambience, with only the soccer game having any real life – but it didn’t feel real, it just felt like generic children playing sounds. The bar scene had some life in the establishing shot, but I didn’t feel like the voices really matched the vibe or the clientele. And the background voices drop almost entirely after a couple of seconds of dialogue from the main characters, which is really jarring. In the RebeLoop version there was much more life, and specifics were noticable.

2. A really great example is the cut at 2:50 to the group of four in the bar. AI: silent. RebeLoop: alive and matching what is on screen and characters/tone of the scene. I would also note 0:50 as the four women walk away – the laughter/cheeriness of the RebeLoop version is a subtle but great contrast to the look of stress/despair/aloneness of the main character that is totally absent in the AI version.

3. This is a tough call. On one hand, if you’re going full Generative AI (assuming the leads are AI voiced as well?), then stick with it for better or worse, just like how I believe a human crafted project should have no AI elements. But, if you’re trying to create a project at this stage of Generative AI where the results are a piece that need the polish of something like looping, then humans are an absolute must if this is any indication of how AI is going to handle those background pieces.

Kyle K
————

1. I definitely could feel a difference between the two. I felt like the Rebeloop version felt more lively throughout and less “canned” in the places where the background walla really stood out. The voices felt more natural and tailored to each scene and helped ease a bit of that uncanny feeling that came from the AI on screen.

2. The opening montage especially with the kids playing soccer, the students picnicking by the river, and the first four women she approaches as they walk away. The bar scene felt more natural and lively as if there were actual people there rather than a backing track of voices that didn’t truly fit. Though I will say I heard both a “Probst!” upfront and a “Slainte!” later that felt slightly repetitive or forced, but the rest of the bar scene felt great. There’s a “Sit down, please sit down” around 3:20 that stands out to me and feels like such a nice addition I can’t explain.

3. I would suggest creators are going to use AI then they should also use humans when they can, as it feels like it can help to ease the uncanny or unnatural feeling that comes from the generated images on screen. I thought Rebeloop did a great job of matching and filling out the picture in a way that stock sound libraries could not.

Ryan K
————

The sound of people talking in groups under the credits was a bit distracting, actually. I felt like those were lines of duslogue that I was missing.  I liked the children playing and I liked the background in the bar much better.

Elizabeth Z.
————

1. On a very zoomed out level I do think there was a noticeable difference between the two. There were more levels to the human background than to the AI one–it was dynamic and there was more variety of tamber, volume, etc. The fully AI had more of a murmur-y white noise characteristic. However what really distinguished the two was in the details:
2. The people in the background in the AI sound version felt directionless, like there was no context to who they were, what they were doing, or where they were going. This comes across I think especially in the scene in the first half of the short where the group of girls are walking in a row away from the “camera.” The AI has limitations in how realistic or natural it can visually convey these things, and it only gets amplified by being unable to generate truly realistic audio as well. In the AI version, even the way the girls moved their bodies as they walked looked off, and it felt uncanny. In the human version, despite the visual being the same, the inclusion of specific energy delivered by the actors gave a more purposeful sense to what they were doing. Instead of walking into the void, they were walking to maybe the library after school. They walking all wonky like that because they were laughing about something and trying to be sneaky about it, and they maybe weren’t really in a rush to get to their destination. This same dynamic is present throughout, more apparent whenever the background actors are featured more centrally. One issue with generative AI visually is it’s not good at choosing when to add detail or not. It’s kind of like a camera that doesn’t focus on its subject. If you ask AI to fill the background with specific activity it’s going to focus too much on it, and it results in confusing composition. The realistic audio helps balance out the intense detail in the imagery, improving things, but it ultimately still comes off as amateurish and… not good in my opinion.
3. This is tough, because I think that yes, it would be good if AI Creators (I had to refrain from scare quoting that one) would employ people instead of automating *everything* and that using humans for their projects will improve them. There’s also something to be said about how generative AI is already being used for writing, behind the scenes coordination, and other aspects of the production process, and so there’s an argument that generative AI being used for visuals shouldn’t get more scrutiny than those other instances, or even that we’re on an inevitable path towards generative AI being used and so not participating is career suicide. That being said, as an artist there’s a certain scab-like feeling to contributing to a project that uses generative AI for its visuals. Generative AI is inherently plagiaristic. In addition to not hiring humans to create the visual experience of the project, be that a production crew and film actors or a team of animators, generative AI is actively using assets created by real human artists, abstracting them from their contexts, synthesizing them, and then spitting them out as though it was original work. That’s not something I would want to be a part of, and unlike AI generated lines in a script or production documents, it is much easier to draw a line in the sand for something that is outright an AI creation. Of course it gets muddy again when there are bits and pieces of AI generated imagery within an otherwise human product that are hard to spot, but I think as an industry we need to be firm about this where we can. And I mean, look–morals/ethics aside–our looping definitely improved the short film, but we polished a turd. And to belabor that metaphor, no matter how much we dress it up, spray sweet smelling perfume on it, or mix it in with something organic, we’re still telling the audience to look at, smell, touch, and eat shit.

Sorry for the negativity, I’m a luddite about AI to an extent that goes beyond rationality at times. Hope that it’s an interesting viewpoint for y’all to chew on as you do broader reflections on the project. I really wonder what it was like for the engineer to have to mix and what it was like for you as the group director having to compose us to create something viable for what you were given.

Ethan
————

I don’t think there’s any doubt that the Rebelloop version is more lively. You can hear the voices just a bit more distinctly. In the original,  It’s hard to tell if  anything going on. The lip flaps on the cheers doesn’t match on either version. The mouths on the AI doing enough. This is a big difference between AI and humans. If you ask a human to sit there and do a cheers their mouth will do what it would actually do every single time but on an AI who knows what it’ll do. And they weren’t perceptive enough to go back and change it or maybe they did try to and just couldn’t. 

That’s just from a sound perspective, from a visual perspective it looks so completely and totally AI. You can really tell and it throws me off. It’s hard for me to focus on the actual film I’m so distracted by the unnatural characters. They capture some of the subtlety of the performance on the faces, but there’s something about a human being and the thousands of tiny little muscles that go through the face and the forehead that just can’t replicate right now. And there’s something about the eyes I can’t even explain exactly what it is just something about the eyes that aren’t quite alive to me are. The eyes window to the soul, but I just don’t see much there.  

David 

———————-