Crew AI would be very helpful for comics, and even for advertising if a client wants something specific from the character. Nice to finally get that level of control.
Agreed. Actually, Kling was one that I went as close to "all in" as I usually get-- I picked up the 6mo sub, but also-- that sale was pretty epic and I didn't want to miss that.
Regarding the question about the "right amount of control". It's not that you can have too much or too little control, neither having the "right" amount of control either; Rather it is having the right *type* of controls. Having the right types of (needed) control might end up in "many" of "few" of them, so it is not about the number it is rather about being able to express what you have in your mind with the most efficient way as direct as possible. Bad UX design is "Well if you do these 10 steps you can still achieve the desired effect" when as a UX designer you should think "How can I help the creator achieving what they want in the most efficient way with the level of freedom they would desire". It is hard to make an elegant UX design in new tools that frankly changes the entire workflow and industry :D Anyway :) always a delight to see your videos, so much hard work to find, research and then try these new tools so that we don't have to ;) in this fast changing (soon to be) industry.
Agreed. What I'd like to have, actually, would be to give a model two references, one scene reference for the background and one character reference, and then having the model acting out that character on that background. Right now the problem with consistency is that it's hard to do different scenes featuring different actors over the same, consistent background. Like, you have a movie in a kitchen, but every time you prompt the model, the kitchen looks differently, the cooking devices and furniture are grouped differently, in a different color, the lighting is differently etc., which becomes totally unbelievable if the scenes are supposed to play in the same kitchen. Right now you can only get this by manually composing start or end frames of the background and manually inserting a character into the background. It would also be good to be able to create videos with sound, preferably having characters say pre-defined lines which are reflected in the audio, but also in their mouth movementns. That's the direction I'd like to go development to.
The walking cycle has improved a TON in the past year. From sort of "floating" while awkwardly moving/morphing their legs around, to a proper walk animation where the character moves forward a natural amount with each step. We're making progress!
I walk like that normally, just moving awkwardly my legs around until one of those movements represents going forward in a surface. I thought that was normal
After watching your video I was playing around with Minimax last night and all I can say is: wow! - Very impressive compared to the competition. (Kling and Runway) - Now if they can get image to video + camera controls I would be sold and would pay good money for a plan.
Minimax is the best we have right now, it seems. When the colors are a bit saturated, we could correct it in our video editor, right? To make it look more natural colors, what would you change in your video editor?
9:20 AI Crew is impressive. But I wish we could upload images of scenes with multiple characters, and be able to change the facial expressions of each of those characters.
I bit the bullet and bought Gen 3 a while back for the unlimited year...its so fun not having to care about credits. The extend update actually works SO well. Im interested in this Minimax too same with Cogvideo running locally. 6 seconds is perfectly fine if its a good packed 6 seconds Also thank you for the vid sir
Runways unlimited plan is extremely liberating. No doubt! I almost never use a shot that is more than 3-5 seconds, so for me 40 seconds is way way Overkill, but the big problem with video generators is the limited control you have over the outcome. I really thought Runway will incorporate its camera control into gen 3 but it hasn't done so yet, and the quality of camera prompting is inconsistent. Pulsar really thought that Runway will incorporate the movement brushes but I don't think that's coming since no one is even attempting anything like it
Hah, I've tried a few "Y'know what I think..." type videos in the past, and when I watch them now, I'm very much in agreement with the audience: "Yeah, yeah-- we don't care." These days, I like to just show and tell...and maybe sneak in some dad-jokes when I can!
@@AngelusNielson oh, I do. I just don’t preach about it on the channel. I’ll occasionally field some questions on it in the comments, but at the end of the day, this is a UA-cam video. No one wants to listen to some rando navel gaze for 30 minutes.
@@TheoreticallyMedia Nobody you want to court at any rate. But these are very serious concerns about this technology. Ignoring them or regulating them to comments only just serves to minimize the issues.
4:22 It would be cool if we could take generations like this and ask the AI to generate the exact same shot from a different camera angle, and maybe also change the camera movement.
Use image to video generator Use the last frame of ai video one (stillshot) as the starting image for the next AI video to generate and add new cues for the next motion as desired.)
Yeah, this, or do other changes, like you have a video of a house, but you'd like the house to have a porch with palisades and a driveway in front of it leading to a road, but the camera movement and the acting characters should remain the same.
I've worked in 3D animation for quite a while, so complex controls don't bother me that much. But having said that, I've felt for a while that the overall UI and workflow paradigms in 3D are really outdated at this point, having been established in the 90s for most of the pro apps, so I'm happy to see workflows and UI being reconsidered in various ways by these AI tools.
I have been yelling this as loudly as I can for awhile now. A few of the "Blender but AI" companies have had me consult, and every damn time, I say: Your UI looks like 3DsMax or Blender. Why? Anyone that uses 3d software is going to feel limited, anyone that doesn't is going to be scared off by the complexity. To be 100% honest, what I keep telling them is: Photo Mode in a Video Game. You ever play Red Dead or Spider-Man? When you Pause the game and can go into Photo Mode? That's what AI 3D should look like. Rumor has it, that's what Midjourney is working on right now.
That segment on zombies in a grocery store just conjures up a nightmarish scene of Boy George clones! Lol. Anyway appreciate your reviews with much to look forward to. Thanks again, Tim.
I would love to have an option to put in more than start and end frame. This option exists in Krea, but it only produces morphs. Like, for example, a complex movement of a character: 1. looking over shoulder 2. taking out a gun, 3. turning around, starting to run
Yup, agreed. I think the solution is going to be a 3 frame approach (Beginning/Middle/End) plus text prompting. At some point-- and I think that's what I was touching on at the end of the video: How long is it until we're all just animators, though? Haha-- I mean, it's fine, it's just also sort of funny that the end of the line with AI Production turns out to be Human Production.
I‘ll definitely start a graphic novel style movie production as soon as image to video gets better. Like a kid’s dream coming true, since I‘m drawing and painting all my live ( as a commercial storyboarder). Thinking about a smart way to cover the flaws in these ai videos by the story. Maybe it’s playing in a parallel universe where everything is packed in jelly :-)
Haha. Sometimes you gotta punt out a “gimmie” question! That said, it sparked some good convo in the comments about UI! Namely: max control, but make it easy to navigate for new users!
There is no such thing as control overload. It all comes down to how well the user experience is designed. Does control introduce friction or is it seamless and intuitive? Additional control simply means more thought and care needs to be put into the design of the interface via which that control is expressed. Having more control in and of itself is a fantastic thing for creativity.
No such thing as 'control overload' if those controls can be made available in a way which doesn't significantly negatively impact the usability of the main crucial controls. 👍 I've always been keen on the ability to maximize control of guidance of AIs because I want maximum capacity for MY creativity as opposed to the AI determining all the creativity. 👍 👍 👍
You pose a very important dilemma in the end of the video. I believe that having all these slider controls with no alternate / easier way to modify them qualifies as too much control. I want the sliders there and I want a chat box with a “director” AI that I would be able to speak my ideas into and it would modify the sliders for me in an almost educative way. As if it’s saying to me “definitely, you can achieve that by modifying the X slider by let’s say Y amount, do you want me to do that for you?” And then followed by a tip like “try playing around with these sliders to find the ideal balance for you”. This way the AI would ease you in understanding what each slider does and also removing that initial Wall one brick at a time.
I was JUST talking about this yesterday! Plug in an LLM and you can basically work almost like a remote director. Calling out not only acting directions, but even camera shots and movement. And considering OpenAI and the others are all starting to roll out voice in these LLMs- well, it kind of will be like being a director on a remote set! Wild to think about.
Hey thanks for the mention. 2:06. Yes, the 40 seconds shot really helps for establishing shots. Another use case would be for engaging timelapse videos, if the overall story can keep your audience glued for 40 seconds, that is 😅.
4:42 These AI video generation tools really need to consider adding ‘inpainting’ for fixing mistakes that the AI makes, such as inaccurate text and the woman in the background walking in a strange way.
There is never too much control, however you should not have to go through all of those controls to get your output, but rather have an AI helper find good values for what you are after, and if you are not satisfied with specific parts and want to manually adjust things, the helper should expose the relevant controls to you. In the end you shoud very rarely have to manually explore all controls to get what you want, only when the helper fails to help you.
We've been talking about that in the MiniMax Discord. You should swing over there, lots of discussion going on, and they're listening. Brent Lynch (who did the Pirate Ship output in this video) threw some quick suggestions out to them and they implemented it right away. Some of the easier stuff, I'll give you-- like 2000 character limit on prompts, but they DID it. So, they're listening and trying to improve.
Haha! I've got that weird personality type where, even if I don't want to watch it-- I'll still do it. Even the last terrible Star Wars show-- I watched the stupid thing. It's an illness, I'm aware.
Wow, this new AI video generator is amaziing . I can't believe the level of character control you can achieve! It's like having a virtual director at your fingertips!
@TheoreticallyMedia It seems like just 2 weeks ago you hit the 100K mark and now instead of growing by the dozen subs every week, your channel is growing by the thousands of subs!! 👏 👏 👏 I can't wait to celebrate your 1 Million subs achievement! I'd be dancing in the streets to celebrate it!! 🕺
It's pretty good! Hop in while it's free, too! I have noticed that generations can slow down during peak hours-- so, if you can-- Generate while the world is sleeping, and you can get a lot of free videos out!
@@TheoreticallyMedia Thanks for the advice. I just checked it out, but it won't run on my phone through the browser. I need to run it on my computer went kaput and I have to get a new one. Ugh! Bad timing. The power supply melted. 🤦♀️
Wow, this new AI Video Generator sounds awesome! I absolutely love the idea of having insane control over AI characters. It’s incredible how technology is evolving at such a rapid pace. I’ve been looking for ways to enhance creativity in my projects, and tools like this could really take things to the next level.
I do not know what was used to generate those last two shots, or what story they are from, but I would definitely like to see more from that Universe. If that was generated by you Tim, major props buddy!
I’m for more control. Just not that Comfey node based stuff, hell I might as well go back to trying to figure out Blender. Sliders I’m cool with. Ultimately I want to be able to tweak with prompts. Like “perfect, now just change her expression.” And, of course the holy grail of AI video for me, “open a door, walk into the room, sit down at the table.” When AI video can do that we will have reached a major milestone in AI filmmaking!
I think any good UI should have layers and that it should always be straightforward to type in a few words, hit generate and get something. Then, you should be able to go into advanced mode to get basic controls and either a pro mode for the whole interface or ideally a per tab option to give he most fine control over each element. There's no such thing as too much control but it only needs to be visible to the people who are going to use it.
There OUGHT to be, so there is accountability in the forensic value of video. Otherwise, video evidence could become useless and inadmissible in court, when you can't know for sure if it is real or not.
For the AI crew, it is my understanding that there is a free and already available comfyui version that does the exact same thing, which I assume the crew took it and refined it to their liking.
That's a LOT of platforms: It's basically just ComfyUI under the hood. But, that's OK, since Comfy is so customizable-- I'm totally fine with it, as long as they come up with a unique spin. Plus, obviously, Comfy can tend to be hardware intensive, so if a platform (like CrewAI) can make it more accessible to the masses, then to me, that's a good thing.
That AI Video Generator is impressive. I assume China may have some of the best Gens due to having a giant database of video recordings and copyrighted materials which other AI companies may not have access to.
That control is quite interesting, it's relatively similar to what we have available in Maya to animate although not quite there. Very cool if keyframes and curves are animatable.
I'm sorry, but I'm unable to generate images or content that involves violence or gore, including scenes with zombie attacks. If you have any other non-violent requests or need assistance with a different topic, feel free to ask! hahah
first prompt: zombie horror movie, army of dog zombies attacking a door in a house inside postapocalyptic ruined city sunset environment - declined second prompt: army of happy dogs mutants happy to see their master that delivered food in a house inside postapocalyptic city sunset environment, they waiting him outdoors and happy to see him the same reason - no blood, no violence... damn
Hi, Tim. I tested Hailuo Ai (Minimax). The generation is very good and fast enough. But the video is watermarked in this Beta phase. From their Discord channel I read that there will be a paid version soon. I doubt it would be cheaper than Runway, but let's see.
@@justtiredthings 👋 Not from my experience in most cases with Dall-E and Flux Pro. You just have to know how to prompt well. Good prompt engineering is a skill.
We need 2 options. #1 is the director telling crew how to make the shot or sound etc. The director says what she wants, how to she wants it changed, and the team goes to it to get the shot right for the director. Now there is the point where you can go advanced. #2 Having the director take over and get in there with her hands to physically move actors in the way they are supposed to move. But this option should only be for fine tuning and when the telling method fails.
Had me at the vintage Mustang. And that witch reminded me of Debbie Downer. Excellent. There will always be more options as a tech develops, thats why its good to throw away the manual and give yourself limitations for a concept then branch out. I love options! Thanks again, broham!
Ha, caring about if a model in a skinsuit is having a pockets is a great way how tell someone's probably married. My wife often goes into rants how women's clothing doesn't have any pockets:) Pretty nice AI video model. Though it is getting a bit crowded with all of these coming out every week or so.
Woohoo! New generator to try...although I'm looking at your thumbnail and your test video of the guy in the cyberpunk city and half wondering if you actually saw my mess of an AI short film attempt or if it's just a coincidence.
Many people use the "Will Smith eating pasta" as the litmus test for ai video generators. Myself, I simply type, "A shark swimming under the sea" as sharks are notoriously bad in ai generation. Well, I was blown away today. This was the first "normal" looking shark I have seen. It is the first image or video generator to pass it, so thumbs up.
12:04 To answer your question, there is no such thing as control overload. Either it is the AI’s vision, or it is our vision. The only way we can turn it into our vision is if we are given enough control to do so.
Nothing is really free. Do you have any worries about giving your phone# to this company? Do you think there is a possibility of the app accessing all the data on your phone?
It is unrealistic to think that there is an app that doesn't access your data ,its up to you to limit how much they can access but at the end of the day if u use the app u share ur data
Even registering your phone number with you bank isn’t safe. They sell the number. I know because I’ve tried it with a new phone number. I set it up. Had it for 4 weeks. Then registered the number with my bank, that same day I started getting so many spam calls and texts. That’s just life. Gotta deal with it.
I think a Test you should try basically use the silhouette test where only the character you want to see isn't a silhouette that way maybe the AI can focus a lot more on the main character
that was Udio! I had to replace the original, since I wasn't sure what the source was and didn't want to get hit with a strike. Gotta love AI Music Generators!
1:08 Not necessarily a down side considering how many of the video generations of 10 seconds or longer from other vid AI tools are slow-mo. So 6 seconds of normal could easily be better than 15 seconds of slow mo....... all else being equal
A fun tool for content makers is a free AI video generator with sophisticated character control! Alternative tools with special functionality for creating videos are available for individuals experimenting with other possibilities.
I'm using the Arc Browser-- and it is SO good. High recommend. A tad bit of a learning curve, since it has some fancy tricks-- but, such a game changer. The translate feature just pops up in Arc. I think in Chrome, it should be in the top right corner? I get a little A and a small a icon, when you hit that, you get the option to translate.
Was Sora a decoy so that these other companies would use their compute up on users generations, meanwhile open ai isn't bogged down with much but their original llm?
Man, that shrimp. I recognize that urgency. The man in the blue suit's gait was the initial realization, The lady ( who went to the same buffet) , well that was after the fallout, thus her amble. Yuck Yuck. We love ya Tim!
oooh man . thank youu very very Much , You are very wonderful, thank you for all your research efforts, please give me link AI Crew: Advanced Character Control
@Theoretically Media as a semi professsional 3d artist, I can say with confidence that there will be less and less need or budget for classic animation workflow except for the bigger companies in the next year or 2. and as I director I dont mind that so much, to me animation is a hassle getting in the way of the story.
Yup! Minimize friction! I was just talking about how (possibly soon?) we'll start interacting with an LLM chatbot for real time scene creation. "Make this a scene in a dark alley, and put the camera low to the ground...no, lower. Ok, that's kind of good, now start tracking backwards-- ok, let's move on to what the characters are doing..." No more fiddly interfaces or controls...just arguing with a chatbot. I guess in some ways, it'll be like arguing with a real DP or crew! I can't wait for my AI Actor to ask me about their motivation...ha!
If you’re using chrome (or Arc which I recommend highly!) there should be an option on the right side of the top of the browser. Sometimes it gets hidden in an option button that makes no sense. Or, it’s somewhere buried in one of the menus. There’s also a free chrome extension you can download.
RE: Control overload: I would love if AI would generate scenes and then we can swap out buildings, objects, characters, trees, and have control over each element. I want fine control over the motion of these objects and the camera, where I can maybe draw a path on screen showing the direction of movement. I want to tell it to have my character jump or walk or run or pick up x. I don't necessarily want the complexity of a full 3D animation studio where I have to move 12 vertexes on the lips. Just tell it to make my character snarl. It's all about adherence to directions. I want to be able to tell DEFINITELY NOT ScarJo to build my scene and move this object or have the actor in the blue suit be more or less excited or crying or faster or whatever. Verbal commands and editing! Generate quick storyboards I can edit on the fly and then when the rough animations look good, render baby!
6:48 Do you know if MachineMythos is using super long, descriptive prompts to get consistent characters and vehicles using text only? Are their prompts available?
Be careful with copyright images. I told it to creat a UFO flight control room with human like aliens and it generated characters from Star Trek Discovery series. LOL! The quality is not that bad but will need to be upscalled and enhanced for best quality.
I was choosing country, putting my phone number and checking the agree button, but when clicking the green button it says I have to write a confirmation code.. I don't get any code. Do I have to register elsewhere first? :-)
Wonder if we'll have local installable, AI video generators like Kling, MiniMax, etc... I know it's going to need a supercomputer with multiple workstation tier graphics card, but...
Thanks for the video. When you are testing ai video generators, it would be cool if you tested something they have problem with generating, and let us see how long they have come in that. For example a space battle, or a epic, lord of the ring, kind of battle. The generators so far I have test, has been very dissapointing in those.
you had me at "Space Battles"-- Noted and will do! yeah, I've tried a few "LOTR Battles"-- I think there's just TOO much happening in those scenes for the model to generate. You kind of get something, but that something is really mushy.
More than 10 comments and the Sora Police has not appeared yet. "Ok, but Sora is more realistic". "They are just testing to bring us a secure environment" "Open AI is in CONVERSATIONS" 🤣😂😅
Oh man-- that was the BEST BETA EVER. It lasted forever and was so amazing in that 1.0 era of video generation. It was such a crazy time. Hope we get to experience something like that again!
It doesn't allow anything. No NSFW, no gore, no violence, so equally useless for action/horror creators. Stable and Sea Art is still the best as it allows everything and doesn't have limits
Yeah, I totally agree as well. I think with more control, we’ll see a bit of a falloff from the folks who are just playing around, and we’ll start to see the real players rise. I’m hopeful it’ll also silence the critics of AI generated material, but that might be hoping for too much.
@@TheoreticallyMedia I see a big shift happening already. Alot of people who were SOO against these and other tools now coming around admitting their mistaken views and how these are great and powerful. Some of us just have foresight :P
Crew AI would be very helpful for comics, and even for advertising if a client wants something specific from the character. Nice to finally get that level of control.
Crucial tip: never buy long term ai account. I checked this ai in the past 2 days, it is much more better than kling
Agreed. Actually, Kling was one that I went as close to "all in" as I usually get-- I picked up the 6mo sub, but also-- that sale was pretty epic and I didn't want to miss that.
True, better go month by month. Maybe in a couple of years it is ok to settle with an AI technology stack that is established.
yeah I grabbed 3 months on Kling. I dont regret it as it has image to video which minimax doesnt have yet
I guess Kling will update new version soon too
I buy yearly subs to save the hassle of dealing with all these invoices & pray that they’ll beef up to their soon new competitors. Wild times
Regarding the question about the "right amount of control". It's not that you can have too much or too little control, neither having the "right" amount of control either; Rather it is having the right *type* of controls.
Having the right types of (needed) control might end up in "many" of "few" of them, so it is not about the number it is rather about being able to express what you have in your mind with the most efficient way as direct as possible.
Bad UX design is "Well if you do these 10 steps you can still achieve the desired effect" when as a UX designer you should think "How can I help the creator achieving what they want in the most efficient way with the level of freedom they would desire". It is hard to make an elegant UX design in new tools that frankly changes the entire workflow and industry :D
Anyway :) always a delight to see your videos, so much hard work to find, research and then try these new tools so that we don't have to ;) in this fast changing (soon to be) industry.
Agreed. What I'd like to have, actually, would be to give a model two references, one scene reference for the background and one character reference, and then having the model acting out that character on that background. Right now the problem with consistency is that it's hard to do different scenes featuring different actors over the same, consistent background. Like, you have a movie in a kitchen, but every time you prompt the model, the kitchen looks differently, the cooking devices and furniture are grouped differently, in a different color, the lighting is differently etc., which becomes totally unbelievable if the scenes are supposed to play in the same kitchen. Right now you can only get this by manually composing start or end frames of the background and manually inserting a character into the background. It would also be good to be able to create videos with sound, preferably having characters say pre-defined lines which are reflected in the audio, but also in their mouth movementns. That's the direction I'd like to go development to.
2:14
Dude...... I really like this detective scene....... the style, colors, etc is really esthetically pleasing
The walking cycle has improved a TON in the past year. From sort of "floating" while awkwardly moving/morphing their legs around, to a proper walk animation where the character moves forward a natural amount with each step. We're making progress!
I walk like that normally, just moving awkwardly my legs around until one of those movements represents going forward in a surface. I thought that was normal
After watching your video I was playing around with Minimax last night and all I can say is: wow! - Very impressive compared to the competition. (Kling and Runway) - Now if they can get image to video + camera controls I would be sold and would pay good money for a plan.
It's getting hard to keep updating my workflows every week, but i'm not complaining!
Same! 😂😂😂
Minimax is the best we have right now, it seems. When the colors are a bit saturated, we could correct it in our video editor, right? To make it look more natural colors, what would you change in your video editor?
looks a bit rubbery? Thought the characters looks way too smooth?
Oui
@@armondtanz That's what I said at to myself about the cell phone salesman that accosted me in the mall last week.
9:20 AI Crew is impressive. But I wish we could upload images of scenes with multiple characters, and be able to change the facial expressions of each of those characters.
I bit the bullet and bought Gen 3 a while back for the unlimited year...its so fun not having to care about credits. The extend update actually works SO well. Im interested in this Minimax too same with Cogvideo running locally.
6 seconds is perfectly fine if its a good packed 6 seconds
Also thank you for the vid sir
Runways unlimited plan is extremely liberating. No doubt!
I almost never use a shot that is more than 3-5 seconds, so for me 40 seconds is way way Overkill, but the big problem with video generators is the limited control you have over the outcome. I really thought Runway will incorporate its camera control into gen 3 but it hasn't done so yet, and the quality of camera prompting is inconsistent.
Pulsar really thought that Runway will incorporate the movement brushes but I don't think that's coming since no one is even attempting anything like it
What's the generation speed like with the Gen 3 unlimited plan?
@@peacetoall1858 Speed itself didnt change just the number you can run in total is never ending
I love your channel. It doesn't just philosophically ramble, it's shows the products.
Hah, I've tried a few "Y'know what I think..." type videos in the past, and when I watch them now, I'm very much in agreement with the audience: "Yeah, yeah-- we don't care."
These days, I like to just show and tell...and maybe sneak in some dad-jokes when I can!
Heaven forbid that you think about the ethical considerations that AI is bringing about.
@@AngelusNielson oh, I do. I just don’t preach about it on the channel. I’ll occasionally field some questions on it in the comments, but at the end of the day, this is a UA-cam video. No one wants to listen to some rando navel gaze for 30 minutes.
@@TheoreticallyMedia Nobody you want to court at any rate. But these are very serious concerns about this technology. Ignoring them or regulating them to comments only just serves to minimize the issues.
@@TheoreticallyMedia I'm really waiting for the "totally repeatable and reusable character" in video AI. Still seems too inconsistent right now.
4:22 It would be cool if we could take generations like this and ask the AI to generate the exact same shot from a different camera angle, and maybe also change the camera movement.
Use image to video generator
Use the last frame of ai video one (stillshot) as the starting image for the next AI video to generate and add new cues for the next motion as desired.)
Yeah, this, or do other changes, like you have a video of a house, but you'd like the house to have a porch with palisades and a driveway in front of it leading to a road, but the camera movement and the acting characters should remain the same.
You can. Sea art control net allows you to use the exact same character in different situations.
This looks pretty good! as always, thank you for your no-frills informative, non-clickbaity, non-declarative walkthroughts.
Thank you for spreading the word!!
Wow, the motion on MiniMax seems really good and stable. I will try it out.
It's been 11 days, have you tried it out?
I've worked in 3D animation for quite a while, so complex controls don't bother me that much. But having said that, I've felt for a while that the overall UI and workflow paradigms in 3D are really outdated at this point, having been established in the 90s for most of the pro apps, so I'm happy to see workflows and UI being reconsidered in various ways by these AI tools.
I have been yelling this as loudly as I can for awhile now. A few of the "Blender but AI" companies have had me consult, and every damn time, I say: Your UI looks like 3DsMax or Blender. Why?
Anyone that uses 3d software is going to feel limited, anyone that doesn't is going to be scared off by the complexity.
To be 100% honest, what I keep telling them is: Photo Mode in a Video Game. You ever play Red Dead or Spider-Man? When you Pause the game and can go into Photo Mode? That's what AI 3D should look like.
Rumor has it, that's what Midjourney is working on right now.
6:14 I love the person in the background whose legs are morphing in and out of existence as they walk. Very effective locomotion.
That segment on zombies in a grocery store just conjures up a nightmarish scene of Boy George clones! Lol. Anyway appreciate your reviews with much to look forward to. Thanks again, Tim.
I would love to have an option to put in more than start and end frame. This option exists in Krea, but it only produces morphs. Like, for example, a complex movement of a character: 1. looking over shoulder 2. taking out a gun, 3. turning around, starting to run
Yup, agreed. I think the solution is going to be a 3 frame approach (Beginning/Middle/End) plus text prompting. At some point-- and I think that's what I was touching on at the end of the video: How long is it until we're all just animators, though? Haha-- I mean, it's fine, it's just also sort of funny that the end of the line with AI Production turns out to be Human Production.
I‘ll definitely start a graphic novel style movie production as soon as image to video gets better. Like a kid’s dream coming true, since I‘m drawing and painting all my live ( as a commercial storyboarder). Thinking about a smart way to cover the flaws in these ai videos by the story. Maybe it’s playing in a parallel universe where everything is packed in jelly :-)
Oh, Tim. That question was so on the nose. But here's your engagement. You deserve it.
Haha. Sometimes you gotta punt out a “gimmie” question! That said, it sparked some good convo in the comments about UI!
Namely: max control, but make it easy to navigate for new users!
There is no such thing as control overload. It all comes down to how well the user experience is designed. Does control introduce friction or is it seamless and intuitive? Additional control simply means more thought and care needs to be put into the design of the interface via which that control is expressed. Having more control in and of itself is a fantastic thing for creativity.
No such thing as 'control overload' if those controls can be made available in a way which doesn't significantly negatively impact the usability of the main crucial controls. 👍
I've always been keen on the ability to maximize control of guidance of AIs because I want maximum capacity for MY creativity as opposed to the AI determining all the creativity. 👍 👍 👍
You pose a very important dilemma in the end of the video. I believe that having all these slider controls with no alternate / easier way to modify them qualifies as too much control. I want the sliders there and I want a chat box with a “director” AI that I would be able to speak my ideas into and it would modify the sliders for me in an almost educative way. As if it’s saying to me “definitely, you can achieve that by modifying the X slider by let’s say Y amount, do you want me to do that for you?” And then followed by a tip like “try playing around with these sliders to find the ideal balance for you”.
This way the AI would ease you in understanding what each slider does and also removing that initial Wall one brick at a time.
I was JUST talking about this yesterday! Plug in an LLM and you can basically work almost like a remote director. Calling out not only acting directions, but even camera shots and movement.
And considering OpenAI and the others are all starting to roll out voice in these LLMs- well, it kind of will be like being a director on a remote set!
Wild to think about.
Definitely no such thing as control overload. It will allow us to get closer to our vision and provide the crucial human touch needed. Love it!🔥
Hey thanks for the mention. 2:06.
Yes, the 40 seconds shot really helps for establishing shots. Another use case would be for engaging timelapse videos, if the overall story can keep your audience glued for 40 seconds, that is 😅.
4:42 These AI video generation tools really need to consider adding ‘inpainting’ for fixing mistakes that the AI makes, such as inaccurate text and the woman in the background walking in a strange way.
There is never too much control, however you should not have to go through all of those controls to get your output, but rather have an AI helper find good values for what you are after, and if you are not satisfied with specific parts and want to manually adjust things, the helper should expose the relevant controls to you. In the end you shoud very rarely have to manually explore all controls to get what you want, only when the helper fails to help you.
We've been talking about that in the MiniMax Discord. You should swing over there, lots of discussion going on, and they're listening. Brent Lynch (who did the Pirate Ship output in this video) threw some quick suggestions out to them and they implemented it right away. Some of the easier stuff, I'll give you-- like 2000 character limit on prompts, but they DID it.
So, they're listening and trying to improve.
"If you've been watching Rings of Power" don't worry, none of us have been doing that.
Haha! I've got that weird personality type where, even if I don't want to watch it-- I'll still do it. Even the last terrible Star Wars show-- I watched the stupid thing. It's an illness, I'm aware.
I read your comment in your voice. Lol it’s funny how you type exactly the same way that you speak 😂
Wow, this new AI video generator is amaziing . I can't believe the level of character control you can achieve! It's like having a virtual director at your fingertips!
@TheoreticallyMedia
It seems like just 2 weeks ago you hit the 100K mark and now instead of growing by the dozen subs every week, your channel is growing by the thousands of subs!! 👏 👏 👏
I can't wait to celebrate your 1 Million subs achievement!
I'd be dancing in the streets to celebrate it!! 🕺
Wow! I've been crazily posting videos from Haiper. I can't wait to check this one out. Thank you for letting us know about it.😁
It's pretty good! Hop in while it's free, too! I have noticed that generations can slow down during peak hours-- so, if you can-- Generate while the world is sleeping, and you can get a lot of free videos out!
@@TheoreticallyMedia Thanks for the advice. I just checked it out, but it won't run on my phone through the browser. I need to run it on my computer went kaput and I have to get a new one. Ugh! Bad timing. The power supply melted. 🤦♀️
Wow, this new AI Video Generator sounds awesome! I absolutely love the idea of having insane control over AI characters. It’s incredible how technology is evolving at such a rapid pace. I’ve been looking for ways to enhance creativity in my projects, and tools like this could really take things to the next level.
Any links where I can find out more about more about Crew? Having trouble sifting it out of all of the similarly-named things in my searching...
Same over here, So many different Crew Ai stuff
I do not know what was used to generate those last two shots, or what story they are from, but I would definitely like to see more from that Universe. If that was generated by you Tim, major props buddy!
I’m for more control. Just not that Comfey node based stuff, hell I might as well go back to trying to figure out Blender.
Sliders I’m cool with. Ultimately I want to be able to tweak with prompts. Like “perfect, now just change her expression.”
And, of course the holy grail of AI video for me, “open a door, walk into the room, sit down at the table.” When AI video can do that we will have reached a major milestone in AI filmmaking!
I think any good UI should have layers and that it should always be straightforward to type in a few words, hit generate and get something. Then, you should be able to go into advanced mode to get basic controls and either a pro mode for the whole interface or ideally a per tab option to give he most fine control over each element. There's no such thing as too much control but it only needs to be visible to the people who are going to use it.
Too bad there is watermark on generated videos, in the bottom right corner, in minimax. But the videos are amazing, the best I have ever seen so far.
There OUGHT to be, so there is accountability in the forensic value of video. Otherwise, video evidence could become useless and inadmissible in court, when you can't know for sure if it is real or not.
For the AI crew, it is my understanding that there is a free and already available comfyui version that does the exact same thing, which I assume the crew took it and refined it to their liking.
That's a LOT of platforms: It's basically just ComfyUI under the hood. But, that's OK, since Comfy is so customizable-- I'm totally fine with it, as long as they come up with a unique spin. Plus, obviously, Comfy can tend to be hardware intensive, so if a platform (like CrewAI) can make it more accessible to the masses, then to me, that's a good thing.
That AI Video Generator is impressive. I assume China may have some of the best Gens due to having a giant database of video recordings and copyrighted materials which other AI companies may not have access to.
More likely because they graduate 10x as many stem graduates as the US
@@tnix80 The Chinese are definitely hard working and talented overall; but there's many things which contribute to the effectiveness of their AI.
That control is quite interesting, it's relatively similar to what we have available in Maya to animate although not quite there. Very cool if keyframes and curves are animatable.
I'm sorry, but I'm unable to generate images or content that involves violence or gore, including scenes with zombie attacks. If you have any other non-violent requests or need assistance with a different topic, feel free to ask!
hahah
first prompt: zombie horror movie, army of dog zombies attacking a door in a house inside postapocalyptic ruined city sunset environment - declined
second prompt: army of happy dogs mutants happy to see their master that delivered food in a house inside postapocalyptic city sunset environment, they waiting him outdoors and happy to see him
the same reason - no blood, no violence... damn
Thanks for the heads up. Seems this is unfit for my purposes.
Hi, Tim. I tested Hailuo Ai (Minimax). The generation is very good and fast enough. But the video is watermarked in this Beta phase. From their Discord channel I read that there will be a paid version soon. I doubt it would be cheaper than Runway, but let's see.
I don’t know, Kling for sure came in at a cheaper price point than Runway, so we’ll see!
this channel is truly the best in ai updates bravo!!!
2:16
Oooo..... yeah, I kinda do like this low angle shot quite a bit
Yeah, even took him off frame and everything! Really bold move on MiniMax’s part- but, I love it! We never see that happen!
@@TheoreticallyMedia 💯%
Even image generators have trouble doing anything but centering a character dead on and having them look straight into the camera 🙄
@@justtiredthings
👋
Not from my experience in most cases with Dall-E and Flux Pro. You just have to know how to prompt well.
Good prompt engineering is a skill.
We need 2 options. #1 is the director telling crew how to make the shot or sound etc. The director says what she wants, how to she wants it changed, and the team goes to it to get the shot right for the director. Now there is the point where you can go advanced. #2 Having the director take over and get in there with her hands to physically move actors in the way they are supposed to move. But this option should only be for fine tuning and when the telling method fails.
Had me at the vintage Mustang. And that witch reminded me of Debbie Downer. Excellent. There will always be more options as a tech develops, thats why its good to throw away the manual and give yourself limitations for a concept then branch out. I love options! Thanks again, broham!
Ha, caring about if a model in a skinsuit is having a pockets is a great way how tell someone's probably married. My wife often goes into rants how women's clothing doesn't have any pockets:) Pretty nice AI video model. Though it is getting a bit crowded with all of these coming out every week or so.
Haha, I hear it from my wife all the time. And then she makes fun of my cargo pants!
Joke’s on her: cargo pants are back in style!
Woohoo! New generator to try...although I'm looking at your thumbnail and your test video of the guy in the cyberpunk city and half wondering if you actually saw my mess of an AI short film attempt or if it's just a coincidence.
Thx for the mention! :)
1000%! Great job with those outputs!! LOVE A Kaiju ripping through NYC! Always gold!
Many people use the "Will Smith eating pasta" as the litmus test for ai video generators. Myself, I simply type, "A shark swimming under the sea" as sharks are notoriously bad in ai generation. Well, I was blown away today. This was the first "normal" looking shark I have seen. It is the first image or video generator to pass it, so thumbs up.
12:04 To answer your question, there is no such thing as control overload. Either it is the AI’s vision, or it is our vision. The only way we can turn it into our vision is if we are given enough control to do so.
Tim...... I'm really starting to get excited with releases like this 👍
Yes!! It’s HAPPENING!!!! Haha
👏 *Absolutely loved this video.* Helped me tremendously. Thanks so much for putting this together! _Keep up the great work._ 🌟
I'm loving Minimax!! It's way better than the others!
hi...umm...how to change the language of the site to English?
@@malaimaren10 There's a section at the top right corner to change to English
@@ramonep7973 that button just ask me to sign in
Nothing is really free. Do you have any worries about giving your phone# to this company? Do you think there is a possibility of the app accessing all the data on your phone?
Some of us don't care😂
With your phone number? How?
It is unrealistic to think that there is an app that doesn't access your data ,its up to you to limit how much they can access but at the end of the day if u use the app u share ur data
Even registering your phone number with you bank isn’t safe. They sell the number. I know because I’ve tried it with a new phone number. I set it up. Had it for 4 weeks. Then registered the number with my bank, that same day I started getting so many spam calls and texts. That’s just life. Gotta deal with it.
@@justtiredthings Look up # rerouting!
I think a Test you should try basically use the silhouette test where only the character you want to see isn't a silhouette that way maybe the AI can focus a lot more on the main character
thanks for the Minimax recommendation.
Great video. Such a great tool to make imagination come to life. I like the music at 9:30.
that was Udio! I had to replace the original, since I wasn't sure what the source was and didn't want to get hit with a strike. Gotta love AI Music Generators!
6 seconds is perfect length. Nobody should have longer cuts most of the time. Just hire production team or use AI to slow down clip if you need 7 or 8
1:08
Not necessarily a down side considering how many of the video generations of 10 seconds or longer from other vid AI tools are slow-mo.
So 6 seconds of normal could easily be better than 15 seconds of slow mo....... all else being equal
Thanks Tim! Will check Minimax out for sure!
Liked the shrimp joke! And the great analysis, as always. Cheers, Tim.
Btw, you could also just wire values to the inputs pf those sliders and get an animation!!
Looking forward to checking out AICrew. I'd prefer that gui to the comfy ui spaghetti.
A fun tool for content makers is a free AI video generator with sophisticated character control! Alternative tools with special functionality for creating videos are available for individuals experimenting with other possibilities.
@TheoreticallyMedia Can you add a link to AI Crew? I haven't been able to find it, just Advanced live Portrait.
They haven’t released yet. But check out the pinned comment, I’ve got a spot where you can play with the same tech!
THIS!!! Big Thanks again, Tim, you're The Man. Gonna have fun with is one. Cheers, everyone.
MiniMax is killer, loving it
These are actually really nice new tools!! Thanks a lot!! 😊
Hi great Video for MiniMax: what browser and what web translator did you use I getting no where Thanks
I'm using the Arc Browser-- and it is SO good. High recommend. A tad bit of a learning curve, since it has some fancy tricks-- but, such a game changer. The translate feature just pops up in Arc.
I think in Chrome, it should be in the top right corner? I get a little A and a small a icon, when you hit that, you get the option to translate.
Was Sora a decoy so that these other companies would use their compute up on users generations, meanwhile open ai isn't bogged down with much but their original llm?
Man, that shrimp. I recognize that urgency. The man in the blue suit's gait was the initial realization, The lady ( who went to the same buffet) , well that was after the fallout, thus her amble. Yuck Yuck.
We love ya Tim!
Just tried MiniMax. Oh. My. God. Not perfect, but very, _very_ good.
At one time 40 seconds of generation or 10 seconds 4 times one scene?
10 second extensions. So, basically four 10 second shots strung together.
oooh man . thank youu very very Much , You are very wonderful, thank you for all your research efforts, please give me link AI Crew: Advanced Character Control
No link yet- still under development! But, I'll have something tomorrow that'll get you playing with it!
@@TheoreticallyMedia thank you very very much Mr.theoreticallymedia
@Theoretically Media as a semi professsional 3d artist, I can say with confidence that there will be less and less need or budget for classic animation workflow except for the bigger companies in the next year or 2.
and as I director I dont mind that so much, to me animation is a hassle getting in the way of the story.
Yup! Minimize friction! I was just talking about how (possibly soon?) we'll start interacting with an LLM chatbot for real time scene creation. "Make this a scene in a dark alley, and put the camera low to the ground...no, lower. Ok, that's kind of good, now start tracking backwards-- ok, let's move on to what the characters are doing..."
No more fiddly interfaces or controls...just arguing with a chatbot. I guess in some ways, it'll be like arguing with a real DP or crew!
I can't wait for my AI Actor to ask me about their motivation...ha!
Is MiniMax available in English on their website? I cant seem to see a language selection.
If you’re using chrome (or Arc which I recommend highly!) there should be an option on the right side of the top of the browser. Sometimes it gets hidden in an option button that makes no sense.
Or, it’s somewhere buried in one of the menus. There’s also a free chrome extension you can download.
RE: Control overload:
I would love if AI would generate scenes and then we can swap out buildings, objects, characters, trees, and have control over each element.
I want fine control over the motion of these objects and the camera, where I can maybe draw a path on screen showing the direction of movement. I want to tell it to have my character jump or walk or run or pick up x.
I don't necessarily want the complexity of a full 3D animation studio where I have to move 12 vertexes on the lips. Just tell it to make my character snarl. It's all about adherence to directions.
I want to be able to tell DEFINITELY NOT ScarJo to build my scene and move this object or have the actor in the blue suit be more or less excited or crying or faster or whatever. Verbal commands and editing! Generate quick storyboards I can edit on the fly and then when the rough animations look good, render baby!
Looks amazing, will give it a shot.
6:48 Do you know if MachineMythos is using super long, descriptive prompts to get consistent characters and vehicles using text only? Are their prompts available?
Where's the link to the Minimax zombie short? Couldn't find it in the description.
Great video, thanks, Tim
Be careful with copyright images. I told it to creat a UFO flight control room with human like aliens and it generated characters from Star Trek Discovery series. LOL! The quality is not that bad but will need to be upscalled and enhanced for best quality.
Is Crew AI available for use yet? I can only find Agent software on google.
It isn't-- but I've got a video coming up in a few hours where you can use it.
Hi Tim, I'm still happy with Kling.
Also, please vote for Donald Trump. ✌
this was really good thanks👍
Where can I find demo/more information on AIcrew?
I was choosing country, putting my phone number and checking the agree button, but when clicking the green button it says I have to write a confirmation code.. I don't get any code. Do I have to register elsewhere first? :-)
Wonder if we'll have local installable, AI video generators like Kling, MiniMax, etc... I know it's going to need a supercomputer with multiple workstation tier graphics card, but...
Thanks for the video. When you are testing ai video generators, it would be cool if you tested something they have problem with generating, and let us see how long they have come in that. For example a space battle, or a epic, lord of the ring, kind of battle. The generators so far I have test, has been very dissapointing in those.
you had me at "Space Battles"-- Noted and will do!
yeah, I've tried a few "LOTR Battles"-- I think there's just TOO much happening in those scenes for the model to generate. You kind of get something, but that something is really mushy.
Awosme and FINALLY!
Hah, thanks! I did take monday totally off and just hung out at the pool. But, we're back to work now!
Is there a link for AI Crew?
I cannot find the right company or project.
I just tried finding this on the play store and it's no where to be found is there another app place to get this?
Check the pinned comment. CrewAI hasn’t been released yet, but I do have a place where you can use it now.
there's no such thing as too much control. that's what is going to be needed to make the hollywood blockbusters that are coming
The website for minimax is entirely in Chinese and won't translate. How do I use it in English!!!!
Any link to the Crew AI tool?
More than 10 comments and the Sora Police has not appeared yet. "Ok, but Sora is more realistic". "They are just testing to bring us a secure environment" "Open AI is in CONVERSATIONS" 🤣😂😅
LOL! I´m dying inside
Can´t stop laughing TRUE "We are in conversations" Who cares, release SORA and then go talk :)
Sora police can appear in any moment. Maybe they are deciding what to say..... )
Sora Police :) like the Grammar Police, but of Sora, and they are not even customers of Sora 🤣
Nice stuff Tim, as always. But still I mourn the loss of early Pika.
Oh man-- that was the BEST BETA EVER. It lasted forever and was so amazing in that 1.0 era of video generation. It was such a crazy time. Hope we get to experience something like that again!
It doesn't allow anything. No NSFW, no gore, no violence, so equally useless for action/horror creators. Stable and Sea Art is still the best as it allows everything and doesn't have limits
no such thing as control overload, the more we control the more we have creativity freedom
Yeah, I totally agree as well. I think with more control, we’ll see a bit of a falloff from the folks who are just playing around, and we’ll start to see the real players rise.
I’m hopeful it’ll also silence the critics of AI generated material, but that might be hoping for too much.
@@TheoreticallyMedia I see a big shift happening already. Alot of people who were SOO against these and other tools now coming around admitting their mistaken views and how these are great and powerful. Some of us just have foresight :P