hey everyone so over the last week or so Minimax the free AI video generator that you can use like right now has been popping up with a lot of really cool outputs I spent a pretty good amount of time looking through all of those videos uh plus going through all of the documentation on Minimax plus hanging out in their Discord so I thought this would be a good time to put together a video on best prompt practices and some tips and tricks including how to get consistent characters sort of so we're going to go over
all of that plus I I made a prompt builder for you okay let's dive in and check it out now as a quick FYI for the time being we are just talking about mini Max's video01 model so it is just text to video for right now although Minimax has stated that image to video should be coming within a few days to a week also I found out that Minimax is just six people yeah that was surprising lastly in terms of updates the character prompt length has been taken up to 2,000 characters which I think it
now holds the record for like the longest promt that you can write in an AI video generator now how much it listens to that prompt well that's a whole other thing but I do have an interesting finding that we'll go over in a little bit so kicking off with basic prompt structuring uh what I basically did was went through all of the documentation on minia Max yes it is in Mandarin and no I do not read Chinese so all of this did have to be run through a translator uh total side note I tried running
it through Gemini just because it has the much larger context window and it was surprisingly inaccurate so yeah I don't know uh another airball from Google for me now according to the documentation the very basic prompt formula is a subject plus a scene or a space plus motion or changes main visual subject of course would be any like person or animal or car scene or space is your location anything from the mundane like a forest all the way up to like a space station on Mars motion and changes refers to any action that your character
is taking or any changes that happen in the location environment say like you're in a warehouse and there's an explosion that basic prompt is very much Illustrated in our you know usual man in a blue business suit walking down a busy City street that's why I use that prompt as a starter um again I am really impressed with mini Max's commitment to background characters uh considering that we have this couple here they walk behind our man in the blue business suit and continue on that's something that a lot of AI video generators do not do
as soon as that couple pass behind him they would essentially vanish so uh yeah again well done Minimax in terms of what Minimax calls out as an advanced prompt formula uh this is where we essentially get into the basic prompt formula but we also add in camera movement camera movement obviously refers to anything from like panning in to zooming or handheld to illustrate all of that coming together The Prompt here is the deep blue night merges with the city below as the man gazes at the shimmering lights from a rooftop the camera starts wide zooming
in slowly highlighting the city's flicker Against The Dark Knight Sky panning to emphas emphasize his Solitude I don't know that prompt kind of reads like radio head lyrics doesn't it obviously it did get the starts wide and Zooms in backwards that said we the shot is pretty good so we could always just flip it and post if we needed to one thing I did mention in our previous look at Minx is how I'm really impressed with its like shot creativity uh for examp I mean just a very simple shot here like a guy running on
a beach and the fact that you're kind of getting like this like tracking drone shot out of it yeah that is pretty impressive it also has some really good prompt coherence for example in this shot where you know we quickly pan to the right to reveal another character I do appreciate the fact that the minia max documentation actually kind of calls that out with two points one to enhance the dynamic movement of the scene use camera panning to capture wider shots adding layers of motion or I guess in this case characters but they also mentioned
that you can and maybe should even detail the purpose of the camera's movement uh they say to describe the intention behind the camera movement to add more layers of visual storytelling I haven't really harped on that too much but it is something that I definitely see in a lot of AI videos like a lot of like the camera whipping all around the scene and look I get it it's fun but generally outside of like a you know a guy Richie movie or a fast in the Furious movie you usually don't see you know that level
of Hyper camera movement I'm not going to get into a whole film school thing on camera motivation I have been thinking a lot about it though uh let me know in the comments if you would like to see a a video on that but minax does mention that you know to capture close-up emotions more effectively try using a slow dolly in to gradually focus on a subject's expression so finally bringing you know the basic the intermediate and the advanced prompt Styles together here's our basic shot this is the camera circles around capturing a Female warrior
in silver armor standing vigilantly amongst the dense forest at night I mean it's not bad but it is obviously a kind of a plain Jane shot she is kind of just standing there like in MPC waiting for you to hit the x button in example two which is kind of more of our intermediate prompt formula uh we have the camera starts from the moonlit Forest panning to reveal a Female warrior standing proudly in the ancient ruins the camera continuously circles around slowly approaching capturing the Warrior's profile and focusing on her face the camera shifts perspective
seamlessly integrating the dynamic closeup of her weapons and the armor details it obviously did not catch all of that I mean it is a 6sec shot that was kind of packed with details but I mean you can see where it starts you know with the moonlit shot it doesn't quite get it but it gets that pan down it definitely picks up on you know the idea of moving in to her face as a closeup although we do get this weird bit where like her hand just kind of like spurts out some blood like she's I
guess she's keeping a little Blood Bank in the hilt of her sword which is weird I'm not going to judge it though you know what happens to people who judge the blood sword they end up inside the bloodsword and finally for our more you know Advanced prompt we have the camera starts slowly panning down from the mysterious Sky revealing a Female warrior in silver armor standing alone in the center of a battlefield her eyes reflect The Cold Light of the Moon and in the background shows the ruins of a damaged City with scattered Sparks the
camera begins to descend slowly steadily zooming in and her swords glimmer faintly as they are drawn from their scabbards the background gradually dims as the camera pulls back focusing on the warrior silhouette so obviously it kind of missed the back half of that prompt but you know for the first half I think it did a pretty good job we obviously got the whole like camera descending down into her eyes which are reflecting Moonlight editor Tim here I totally forgot to cover the section on consistent characters you know to be honest it only kind of works
but I did want to go over it so the formula for air quotes consistent character according to minia Max is Hair Plus age and gender plus color of the clothes plus environment plus you know action camera Etc but what I've kind of found Works a little bit better is to work with character Arch types for example here we have this nor detective in an alley uh and then for additional shots uh I just simply kept that Basse prompt but would adjust the actions within that shot and that seemed to lead to at least a somewhat
consistent character and style okay let's let's go back to uh live me so obviously these prompts are getting fairly dense and I should say that you don't necessarily have to prompt with you know kind of ultra long prompts you can you know get really good results out of just using very simple and basic promps like one of my favorites from the community feed which is is weird uh is a weird scary hybrid animal plant scrawling in the dirt practical effects post-apocalyptic movie scene shot on red kodo directed by Andre tarkovski tarov if you are not
aware is a pretty legendary Russian science fiction film director his two I guess like most well-known films are Solaris and stalker they are both very good but they are both very long now because our prompts are basically getting as long as a tarkovsky movie uh one of the things that we can do is build a GPT to do that work for us so basically what I ended up doing was taking all of that documentation sort of formatting it up and then training out a GPT that could write a prompt for you so the minia max
cinemotion prompter is available for you to use uh it is linked down below just giving it the prompt uh give me a western Showdown scene a sheriff versus a man in Black and running that we get this as a prompt which is you know a pretty intense prompt uh I also had it provide a Chinese translation as well since you can run Minimax obviously with a Chinese prompt as well there is some debate going back and forth as to you know better quality in a Chinese prompt versus an English prompt I don't necessarily think so
uh but it's here if you want to the main reason that I have the Chinese translation is that uh because well essentially Chinese is a more efficient language than English is you can actually get a lot more detail in you know beating the 2,000 character limit if you use Chinese admittedly you will probably still have to slightly adjust the prompt here and there like our first output came out with this where you know we have our man in Black here actually this part of the shot is really good but you know obviously our sheriff is
kind of a Hobbit which is a whole different film than I was thinking about going with but nonetheless one that I would watch after doing a couple of quick manual edits to The Prompt uh we did end up with this shot which is much more in line with what I was looking for and lastly since we don't have image to video right now uh one of the things you can do is just simply upload an image to the GPT and it will provide you with a prompt description for that image now important to note you
won't end up obviously with a onetoone of that image it's something that's more inspired by that image but you know it's something of a workaround for the time being until we do get image to video and even then I still think that it's going to have its use case rounding out I do like to Showcase what the like the community is doing with this just so that it kind of shows you the width of what the model is capable of kicking off with Brent Lynch uh two teenage girls in like a McDonald's as a dad
with a daughter that is approaching this age uh I can 100% % guarantee you that the girl on the left yeah those emotions are completely and totally accurate Marco gives us this running shot which there's a little bit of decoherence at the very beginning but I mean that's some pretty crazy motion that's happening there um that's again something that you really only get in text to video this Frame does not exist gives us this claymation Superman short um yeah I'm really glad they did this because this is not an area that I delve into that
much um but it's really cool to see that Minx is capable of doing you know this style big fan of Wallace and grommet back in the day so I am really happy to see that Josh Forwood gives us this uh you know kind of very romantic scene or is it the beginning of a horror scene I mean I guess we're just going to need to wait until we get the extend video function to find out Styles Morales gave us uh Scarlet Johansson as aloy from the game Horizon zero Dawn to be honest she comes out
a little bit more like Elizabeth Olsen probably just skirting the wrath of Scarlet here uh that is a movie that would like to see though over on the Discord zout gives us this scene of a Sasquatch walking through New York I love this and what I what I really really need is extend to happen because what is about to happen here Dr pixels gives us this Like Gladiator lion strolling through a battlefield this shot is chef's kiss and Vex boxal Raptor gives us Jared Leto wearing an its morbin time T-shirt um yeah I've seen all
the memes still have not seen the movie really just don't want to lose an hour and a half half of my life to it so that's the latest and greatest on Minx there is obviously a lot more to come we still have image to video to go and extensions while I do realize that Generations are kind of happening on the slow side right now I mean again it's free so everyone is using it I believe currently Minimax is the only free like completely unlimited AI video generator around and uh from what I've heard you don't
even need to log in Via a phone number anymore you can just go to the site and start generating so let me know in the comments what you think of Mini max if there's any suggestions you guys have on the gbt let me know and I will definitely add them in in the meantime I thank you for watching my name is Tim