This is some experiments I have made with ChatGPT during a discussion. I used one of the sketched I made and try to see what it could generate using sketch + prompt. This is a first try experiment. I uploaded the art of my deviant art:
https://www.deviantart.com/larienna/art/Lightning-Gun-from-sketch-1375685538
I think the results are interesting and I feel responsible for the look of the final product, which is hard to do when using only prompts.
You'll notice that on my Webstore ... All of the Images of the "Dual Dice" (DD) are REAL DICE with an AI Generated "scene". Meaning that we actually took photographs of the dice on white and let the AI determine what would look good give a small "prompt".
I think the pictures came out impressive. But 10 Second Video "sucked" so bad because the AI was not trained on the SHAPES of my dice. So it came up with NOT realistic faces when making a 3D Video (which was cool except not precise enough for me to use...)
As far as your Gun ... It did a great job in rendering it. Reminds me of something you would find in "Fallout" (If you know that franchise). It goes along with the "Pip-Boy 3000" which is like a sophisticated wrist watch with a screen and buttons.
But definitively a GOOD "rendition" of you text sample. Now what I wanted to say... Is that there are platforms that can take that RENDERED IMAGE and turn it into a 3D Object. Never tried, but people are advertising that fact and it could be useful to some people...
Anyhow... Real COOL looking! Cheers.
Also ... In that same thread you used to create the Laser Gun... You could ask the AI to create you an image without any background ("... on a transparent background")... And voila you will have a Laser Gun to be included in your game on a card for a weapon, as part of a logo, etc. etc.
No doubt the current background is "nice" ... But if you want something else for your items or have some kind of "blurry" background ... My point is... Having a transparent background could be helpful for certain reasons.
Just something to HELP you out with. In case you may need this GUN for more than just Deviant Art...
Also be aware that SOMETIMES the AI creates a "checkboard" background... If it does this say: "No I did not want a checkboard background, I wanted a Transparent background." That's happened to me a few times. When I was trying to find Hex Tiles for @X3M.
Anyhow... Just so that you are aware. It'll do it. It just may take a few tries.
Cheers!
Click on this IMAGE to see the full screen transparent background version:
It looks just as good as your "framed" version. You can open a Paint Program that has LAYERS and add a Background Layer and paint it BLACK and then you will have YOUR GUN on a "Black Background".
Enjoy!
Here is the same LASER GUN on a "Black Background". I added a Layer in PaintShop Pro (Corel) and made the background layer BLACK (#000000):
You can download the Transparent version and play around with the Background yourself. I'm just showing you that it IS POSSIBLE and NOT COMPLICATED to do.
Best!
How did you proceed, you used AI to remove the background?
There is still a bit of background under the aiming line.
Like I said, I D/L your image from Deviant Art... Removed the "sketch" with a Black Rectangle... Then Uploaded the V2 Version (no sketch) and then said: "Can you make this gun image on a "transparent background"?"
And it produced me a version with no background.
Then with that version you can use any layered paint program to color the "background" to whatever you want. And then save it to a FLAT image (JPEG or PNG)...
It may not be perfect... But it does a good job. You can try it yourself.
Cheers!
Note #1: Yeah I see what you mean about the glowing light of the aiming ... You can probably take that "no background" image and create a NEW version with an localized correction. But yeah I see what you mean!
Using the AI to correct it... It just "re-designed" the Laser Gun with this:
If you really want to conserve both... You have to fix it manually. What do I mean? Well open the v2 version... Select the blank area and fill it with a color like RED. And then PASTE into the original image (with no background) and then use that selection to DELETE the other layer with the "black lit background". I don't have photoshop on this box... So it would be the best tool to do that manually... Paintshop Pro (Corel) isn't letting me do that "easily".
But it IS possible if you ABSOLUTELY want the original look. But in terms of simplicity ... Maybe you like the version without the glowing cross hairs.
It's up to you... It's not bad in version 2 of the no background IMHO. But it is never the less a "compromise". The whole AI is not perfect sort of perspective. It did FIX the error ... But it went a bit far fixing TOO MUCH.
Like I said, you can use Photoshop to select, paint and the clear that area in the original "No Background" Version. IDK what you have as paint programs... I know it's possible. But I don't want to switch to my Photoshop Box... Just because this is only an "example".
Sometimes it works ... Sometimes it doesn't:
Well I guess in this version ... It WORKED! The only problem is the amount of prompt it takes and images it generates... Makes it HARD to work with ONLY a "Free" Account. If you PAY... You don't have to worry. But I think I have like 5 images FREE and then I need to WAIT 24 Hours.
So this is the IMPROVED version... Tell me if this is COOL or not!
Best.
I've dabbled here and there with the various AIs. And MidJourney is the best to make COMPLEX scenes with little prompt. It invents a lot of stuff and adds to the image to make more complete images than just a prompt with a sketch. ChatGPT on the other hand ONLY does the barest minimum when it comes to illustration.
Like @larienna said... It's hard to get it to do basic stuff like a SOLID "Black" background. That's why I recommend trying to get a "Transparent Background" and then play with Layers in most Paint Programs and add the color or background you want.
But you can also play with the AI and use this image and ask for some kind of OTHER background too... I just did what you asked me: "Fix the cross hairs" and ChatGPT understood.
I had another AI issue with Image generation between NIGHT and DAY. ChatGPT is more understanding at transforming a NIGHT SCENE to a DAY SCENE. It understands what are "lights" and the "moon" and knows how these elements are not part of a DAY SCENE. MidJourney can't do this AT ALL. And ChatGPT shows the most "intelligence" in KNOWING what elements are NOT part of a DAY SCENE.
I've tried DAY to NIGHT ... And that's very HARD. It's much easier to work from NIGHT to DAY (in the opposite direction) because elements like lights and so forth are not part of the DAY SCENE and I would say that it's about 90% there when comparing the AIs.
I think if ChatGPT continues it will be the BEST image renderer because it UNDERSTANDS the elements of "composition" whereas "MidJourney" just doesn't and make more "fancy" illustrations sometimes taking too many liberties. When you go from NIGHT to DAY SCENES the buildings should look ALIKE. And that's something that I've tested in the past and found that ChatGPT to be the MOST effective of understanding HOW(?) to render between those two times of day.
That just some of my exploratory "journey" as I've tried different AIs for different purposes.
Enjoy the AIs and what matters the most: Having FUN and using the tools that are available to help get you the results you WANT.
Sincerely.
I like the Cross Hairs in Golden color... Not glowing blue. I think it makes more sense to be in Gold than glowing blue. So I like version 2. But if you want to stay true to your sketch ... version 3 is it.
Like I said... For me personally... I prefer the Golden cross hairs. So I think ChatGPT went a step further in version 2... And I think that its "mistake" is what looks best.
Again that's just my personal choice. Like I said version 3 is exactly as YOU had imagined it.
Cheers!
With a PAID Subscription for 1-Month to OpenAI's ChatGPT ... You lift the LIMIT of images that it's premium service can generate. With the FREE Account, I can only have like 5 images before it tells me that it cannot generate my "prompt". So that's the only caveat. But hey, MidJourney has never been FREE... Or maybe initially it was FREE for a certain amount of images... Now you MUST PAY for the use of MidJourney AI Bot that creates you the images that you need.
ChatGPT still allows you to do some FREE stuff... But again limited in the way it handles the prompts and "refining". ChatGPT understands and react to the "prompt" in comparison with other images and knows "composition".
So you can still experiment with the FREE Account. It's just that you don't have MANY "prompts" to get the image just right. I've used 4 and I believe that I only have one (1) image left before it tells me I need to WAIT 24-Hours before my limit is restored.
But if you get more heavily involved with image generation via sketches... I'm guessing that you'll probably use the PAID service for 1-Month and just do binge editing of all of those sketches in one (1) Month and get the most out of ChatGPT the (paid) version.
In any event... I still managed to EDIT your Laser Gun and get TWO (2) Versions that look pretty decent: version 2 and version 3.
It's your choice which one you prefer. To me the "Cross Hairs" should not be energy. Energy should be left for the gun and "shooting". The aim should be a sort of target used to AIM. That's why I prefer the "errored" version 2 because (to me) it makes more sense.
Again just my opinion.
Sincerely.
I thought of a way to have a hexagon grid that is seemless. And with ridges going through the hexagon. I could whip up some sketches and see what the AI can make of these.
I know you showed me a larger area with trees and rocks and roads in an INVADER MAT... You could upload that image and ask ChatGPT to create a NEW scene with Palm Trees instead and a different path for the road, etc. etc. And it MAY come up with a new scene. No guarantees, I'm just suggesting it... Because it CAN react to an image and create a NEW scene from it.
So maybe on the FREE Account you can TRY that. Like I said, it will only allow you to generate four (4) or five (5) FREE images a DAY. Then you have to wait 24-Hours.
But maybe give it a TRY and see... I know you already had a scene. Maybe the AI can re-design it with the same kind of theme or different trees, etc. etc.
Best!
I took a sample image you created with a PATH and some Hexes, the source file was this:
And here is the finally RENDERED version, given more REALISM, the output file is this:
Pretty decent if you ASK me... Let me know what you think... If you like this I can explain how to get it from the SOURCE file. The OUTPUT file looks much better IMHO.
Again, let me know!
Note #1: That took TWO (2) Images to create. I can't seem to get away to ONE (1) Image because the first render looked very dark and the details were hard to see. This is much more easier on the eyes and the detail is still very precise. Enjoy!
We're just asking some "other" questions like WHAT(?) can be done with OpenAI's ChatGPT. Again discussing AI Art and what the LLM models can do when it comes to artistic impressions.
I personally think the RENDERED version looks so freaken AWESOME and it mimics the detail of the original image pretty accurately.
And so this means that OpenAI's platform is very impressive in what it can do give a file as input and only a SIMPLE prompt to get the results that we are looking for.
Cheers all.
Note #1: Also please note that the render is not 100% the same. I get that your top tiles were less grass and more sand. It ain't PERFECT but I told @larienna the same thing: if you can agree to the small changes it makes in the image... You'll land up with more solid output even if you had to compromise a bit...
Note #2: I'm all out of CREDITS for today; I have to wait 22-Hours before I can generate more images. It's tricky. I tried to add the SAND but it did not work out as I had planned. Anyhow... You have this sample to work with and see if this is "good enough" for your own needs.
I tried my best... But I cannot generate more images until tomorrow. And BTW I only generated three (3) Images today and it says I'm "out of credits". So I guess it depends on some factors (maybe file size - IDK).
I mean IF you can CREATE layouts like the first image (SOURCE) and I know it involves some effort ... However, with AI, you can render that image you makre and produce an amazingly realistic looking "Area of Play".
And believe me, I know the effort put in to make the SOURCE image. It does take some effort to do that... But if you are willing to make those source images, well then you know you can get an even more impressive result by using AI to render that image.
As I mentioned before, you may need to make SOME "compromises" but nothing too dramatic TBH. Especially for this QUALITY of the "Area of Play".
So I guess my question is: "How do you feel like generated maps as the SOURCE file???" If you are comfortable with that... You can create pretty decent output that looks much more realistic.
Sincerely.
Note #1: When I get more credits (tomorrow), I will TRY to "fix" the top-right side which is supposed to be desert and palm trees. I tried and yup you guessed it: it re-did TOO MANY tiles. I've got to work back some tiles and see IF I can work in the "desert" too... Like I said ATM I am out of "credits" and will wait until tomorrow evening to give it another pass.
We'll see. The current version (#4) is NOT correct and went too far. I won't post it because there is no point in having intermediary files/images when the final ones will look "just right"!
Cheers.
Incredibly awesome!!
Some mistakes though. But the information was not given either.
The edges that contain the rocks, that hexagon itself was supposed to be higher ground. The rocks here were cliffs, but not all though. Not sure why the numbers aren't there. From south to north, right in the middle. I think the swamp was 0, the road 1, the hill 2 or 3, slowly going down to the north, so both the road and hill end up at 0 again.
And the sand in the north is like dessert sand. More like a savannah where there is some green.
I never went into detail in that one. But that is ok.
On a side note. I should make that sketch soon. Not sure if I find the time anytime soon.
In the new method, the hexagons would not have ridges anymore on the sides. But within. So are the ramps. And all art would be isometric?
Because we need shadows too.
Only hexagons would remain. The results: An easier way to make maps. An easier way to show players the properties of a hexagon.
Only last issue, height of a map. Which, with proper map design, would be easily noticable with just some brightness difference of low and high ground.
Please await my sketshes.
For the crosshair, the light crosshair was something he came up with, I do not dislike the idea, but it might be harder to aim with, so maybe the golden crosshair is the best, the sketch only showed a flat line. But its like giving initiative to an artist, it just that AI did it instead. I mean 80% is like the sketch, so I could really say that my mind got put on paper.
Its true that if you design a game and have sketches ready, you could get a month, generate all your art, then stop the subscription. Still I prefer having it installed locally when possible because of the amount of control I could get by saving parameters and prompts in configuration file. I think this is what would make me go all in with art generation: Control on the generator, the use of sketches to match my creativity.
Wow! your map generation is really interesting, I worked so hard for my fallen kingdom maps, it would have been much more awesome to have AI rendered map. You only render the logic of the map, and let AI make it look good. The source map could just flat colors, stamping the map with icons like in wizard king could work too.
I could probably easily generate rooms like Hero Quest or Mansion of Madness.
Still, chatGPT AI said something about licenses that could be used or not for commercial purpose. When we discussed about it, it was not really clear. On another hand my friend has been using claude, nano banana and a few others, and she says that they are all "print on demand" compliant. They are some licences that says that you need to make less than 1 million dollars per year. I like that idea.
And I know, I might get clash from people saying that AI was used. If you look at that image on Deviant art, it has received a comment blocked by admin, so I imagine, it could be about the fact that I used AI. But if you look at the facts, let say I make a thematic game like My steam punk arkham horror game, and generate art by AI, I am not going to make millions of dollars even as a digital game. I would consider it to be an enormous success with just a 10K$ income. But 10K$ is not enough to pay myself and artist considering the amount of art I need. We are not making mainstream games, we are making niche games. It just level our games with the rest of the crowd. Even if 10K is a lot for niche games, its still not enough do to a living. And luckly, I am a programmer, I do not need to pay anybody for that. So questions becomes, ether you get a no art game, or no game at all, or a game with AI art.
I am going to try to use a partial sketch in inkscape since I do not have access to a scanner, and see what it can generate with it. I am looking in making some kind of atari/intellivision art, where the art looks 100 times cooler than the game.
Now that we have AI from sketch. I wonder if we could have something similar for music. Maybe if I could sing some rhythm and melody, then let AI generate the whole song. I can come up with melody, but its tranfering it as playable music the problem. Still a prompt for that would be much more complicated to decribe a music, besides its purpose: "Make a combat music in a steam punk horror game"
Something I just had an "IDEA" for... Because it makes sense that SOME MAPS be in "Winter". Again... No worries... Just "experimentation".
It's a "VERY WHITE" but it looks pretty decent too!
Again... Just an experiment because I wanted to see the END RESULT.
Cheers.
Stable Diffusion is the ONLY AI you can download and run locally provided you have enough GPU Power...
OpenAI's ChatGPT works on monthly plans and Cloud Storage.
The thing to remember is that the LOCAL AI is NOT as powerful as the Cloud AI.
So if I am showing OUTPUTs of ChatGPT for the Hex Terrain Map. Well the truth is probably ChatGPT can generate THAT (What I've shown) and then Stable Diffusion may make something LESS "impressive".
Working with the RIGHT AI is ALSO important. If you insist on LOCAL AI... Well then your Training Data Set is MUCH SMALLER than the set used by "OpenAI's ChatGPT" for example.
But there is ONE (1) CLEAR advantage: you can TRAIN your OWN AI to understand what kind of ART you make. So say you are an ARTIST, you could use "Stable Diffusion" to train ART made by you ... IN YOUR "STYLE". To basically get the ART to DRAW Art like you would normally do.
This is 100% legit because you are TRAINING the AI based on what YOU PRODUCE as ART. And then you can probably generate MORE art for projects that require more art than you actually want to spend time MAKING.
In any event... I prefer the "Golden Crosshairs" too. Looks more natural and for the purpose... "Do you really want to AIM with a LASERED Crosshairs???" I would think NOT TBH!
Anyhow ... I'm glad everyone is ENTHUSED for the AI Art and the MAPS it can generate. I just gave it a TRY... To see what it can make. Anyhow... @X3M can experiment with better prompts and get more realistic results (probably). I just used his image and asked for more REALISM and that's what I got. Pretty neat too.
Cheers all.
Note #1: You can ALSO CONVERT this entire "Grass/Tree" scene into a DESERT SCENE with an OASIS at the bottom:
Please note I had to make a MANUAL EDIT because the AI had removed the RIGHT-MOST "Exit/Entrance" in the Desert Result.
But this could be a more STRATEGIC MAP because you come out from the OASIS and into the DESERT ready to attack whomever is waiting on the other side. This looks pretty decent too.
Enjoy!
Note #1: Also if the "Scenario" or "Mission" is about escorting a Convoy from the TOP-LEFT entrance to the RIGHT-MOST exit, as @X3M suggested the TOP AREA is filled with Mines that force you to go the LONG route. And of course when you reach the 50% marker (where there are hills on both the TOP and BOTTOM) that's when the Guerillas can come out from BOTH SIDES of the OASIS and for the player to battle on two (2) FRONTS.
Could be exciting!
Still, I wonder, how would the AI deal with a cliff?
Well I saw your "sample" cliffs and I was NOT "impressed" by the ART. Yeah it looks "passable" but it doesn't get me excited ... Like the ChatGPT Map did!
I think you would need more PROMPTS to do cliffs or maybe a VALLEY ... I tried to take the Cliffs you showed me, and ChatGPT couldn't do ANYTHING with them. And like I said, I don't really like them. The art is DARK and very "texturized" probably the reason that AI couldn't improve the image.
I just did it to see what the result could look like.
BELIEVE me when I say: "YOUR map yielded the BEST results..."
And I can see COVER from the Forests, on the Mountains or on the Rocky Hills as being ELEVATION points. Like I said to @larienna ... Sometimes you need to make COMPROMISES based on what is POSSIBLE rather than what you WANT to be 100% of your end-result.
So I can see UNITS atop the Rocky Hills firing down on the ROAD below in the "valley". If you look at it that way ... You already have SOME elevation... And some interesting fighting AREAS...
But I cannot do this kind of experimentation FREE... As I need to pay $25 CAD to have more image generation. IDK what the limit is... They don't say... But I'm sure it is MORE than "3" images... Or renderings!
Maybe like 100 images per month... Something like that. Not 100% sure either... It's just an "educated guess" TBH.
I have yet again RUN OUT OF CREDITS (Free AI). I will try when my credits are restored some time after 9:00 PM EST (N.Y. Time Zone). I have some ideas about what to TRY... But I can only generate "3" Images every 24-Hours. So I cannot fully experiment as I would like ... Because of the limitations of my FREE Account.
We shall see... More on that soon enough. Please be patient and we'll see the results (or not!) in about 24-Hours.
Cheers @X3M.
Note #1: Notice that I didn't say that I have a PLAN?! We'll see because there are some logistical differences between what you want as "elevations" and what the AI Generated (Rocky Hills). I will try to marry both types of design and we'll see the end-result for certain.
Really curious how this will all work itself out!
As for cliffs. What if you ask the AI to fill in an empty hexagon grid? And ask the AI to make high and low ground with "ramps"?
Even though, you are not impressed. Please know that those pictures that I provided are 3 decades old now. And only serve as examples in how a cliff looks like that I am talking about. I don't know how else you would call it. Ridge? No, it cannot be a ridge. It doesn't show high to low or vice versa.
But my point being ... You are trying to MERGE two (2) very different things. I don't know if you will understand but one is Hexagons and the other is Squares.
What you call "cliffs" are an Isometric Projections. They are NOT Hexagons.
That's why this exercise of transforming one to the other is ... Interesting!?
I have no clue how it will do this only suffice to say that the AI understands Isometric Views and we'll have to wait and see how it TRANSFORMS Hexagons into an Isometric View.
Again we are ASKING a LOT. It's two (2) DIFFERENT WAYS of making terrain and IMHO I think the result will be very poor. Because you are approaching the same issue (Drawing Maps) with different "technology".
Isometric View are GOOD when you want to SIMULATE 3D. That's why all those RTS games use it: in reality they are rendering a 2D Scene that LOOKS 3D but it is NOT!
And Hex Maps are something totally different in that they are designed such that they can CONNECT with one another... Forming a LARGER Map that can be assembled as part of a campaign not just one "Mission". How often does that happen??? Probably 0% the time in RTS games but in games like Dungeon & Dragons ... It probably happens more often than you would think.
So we are going to see what the AI does with the Rendered Map given a prompt with SOME "instructions" as to HOW to Interpret the Hex Map and TRY to transform it into a 2D Isometric View (Which is meant to depict 3D terrain).
That's why I am saying that you will probably need to either make Isometric Game Tiles OR Hexagon Tiles but NOT BOTH. Because fundamentally they are very different APPROACHES to mapping terrain.
But we'll see... I am "curious" too!
It says that those ISOMETRIC Tiles were made in 2020. Like I said, I personally don't like them. I think they are "Ugly" (no seriously)!
Here's the real issue:
As opposed to the Terrain Maps that I generated: Grass/Forest, Winter and Desert/Oasis ... Those had a much NICER feel to them even if they cannot do some things as you would like to have in an Isometric View.
That may be something you need to COMPROMISE on: either or. So you either go with Hexes (which IMHO look AMAZING!) or you go with Isometric Views and then I would use one of the RTS Games and create the MAPS using their own tool.
I would not waste my time with AI and such... When you already have a DIGITAL tool that allows you to create THOSE kind of maps.
It's OVERKILL to try to make them using AI and Flat Maps. It's quite plainly probably the WRONG method by which to create those maps.
That's my opinion NOW... But it could change depending what happens tonight when I attempt to create a map using the AI. We'll have to wait and see... But my bet ... It doesn't work great. We'll see. I just don't see HOW(?) the AI will convert between two basic shapes to make the Terrain Map.
We shall see...
I think the proper term you are looking for is "Elevation". How the terrain goes up or down depends on the "Elevation". You want some kind of way of representing the "Elevation" and I have some ideas about HOW(?) to do it... We'll just have to wait and see what the AI determines it can DO.
And then take it from there (depending on the results).
Look TBH I hope it does a great job. But I'm not 100% sure that it will ... I know pretty much the prompt I want to use and we'll see how it "takes that" and transforms it to an Isometric View.
Anyhow... 3-Hours to go and I should be able to produce "3" Samples... And we'll see how it goes with the AI. TBD.
Topography is the act of mapping "Elevations" into a 2D map file which produces a 3D Terrain. Here is what I mean:
We maybe could try that if the Isometric conversion fails. It's way more complicated... But the AI might like that level of DETAIL. IDK. It's just an idea to see what you can do to make it more feasible.
Let me know if any of this sounds familiar?!?!
Just moved the NEW ISO Image to comment #31 ... For ease of accessibility.
[Moderated.]
 | Testing AI art from sketch (34) by questccg |
| Health Tracking vs. Threshold Mechanics in Game Design (23) by questccg |
| Damage Tracking vs. Health Tracking in some fictitious Wargame (9) by X3M |
| I published another video/board game (0) by larienna |
 | Mystery Card Game — Thinking of the suits that will work (79) by questccg |
| Dual Dice — The Mobile App (1) by questccg |
| Cooperative Card Game - Thoughts? (2) by Noah McQ |
| Commanders and aces (13) by Noah McQ |
| Do TCGs Have 4 Core Card Effects? (8) by larienna |
| Game design software: protograf 0.5.6 release (0) by gamesbook |
| "Dual Dice" Webstore NOW LIVE! (1) by questccg |
| Dual Dice and Puzzles 2.0 (0) by questccg |
| Textbased game? Lesser game mechanics (88) by X3M |
| Dual Dice — Working on the product (25) by questccg |
| Ludika's offer to board game publishers looking to publish digital versions of their games (0) by heruca |
| I built a web tool for making custom Cards Against Humanity prototypes, would love feedback from designers (1) by X3M |
| Baseball Shuffle (6) by questccg |
| Monster Keep — Renewed optimism (16) by questccg |
| New TTRPG — Working on making an RPG Adventure (9) by questccg |
| protograf 0.5.3. release (0) by gamesbook |
| Alternative to Nandeck? (23) by gamesbook |
| Procedural Investigation Mechanic (8) by Noah McQ |
| Payment for a choice; Paradox (20) by X3M |
| Version 1.29 of nanDECK is available for download (0) by nand |
| Dextrous - prototyping platform (22) by gibwalker |
I am not sure how well it works with transparent background, it cannot generate a plain black background, I tried. One of my friend is playing more with AI because she is setting up an etsy shop, and many images are AI generated.
Apparently, she could generate product video where they put your design on the t-shirt or what ever product you are selling, still its possible that the AI is trained with a limited set of products to make the rendering possible.
Else if you have the product on hand, you can film it. Her problem is that she cannot print every design on every t-shirt of every color to make a video/picture about it.
The AI output is not only awesome, but match my sketch pretty well. I had metallic canister in mind, but I like the glass tube. The blue energy every where makes it look like a lightning gun. Even the electric aiming line is interesting, not something I thought of. So its like if I made a sketch and gave it to an artist. This is more my type of workflow.
Now I have more reasons to sketch again, I don't have to worry that my sketches are perfect or anything, AI will take care of the proportions too. That could make making organic monsters more achievable (ex: a gargoyle). Opening some game theme.