Adobe Firefly vs. Midjourney v5 - Same Prompt Comparison
VloĆŸit
- Äas pĆidĂĄn 1. 06. 2024
- bit.ly/3n6Udx0 đ Learn UI/UX & CSS Today. Use "UI2023" for 23% Off!
designcourse.com/af đ My upcoming "Advanced Frontends" Course
-- Today, we're going to take a look at the same exact prompt issued to both Adobe Firefly and Midjourney v5. Adobe released their invite-only beta access to Firefly, which is their new AI system that will inevitably enhance their products. For now, it's a web-based interface that has a LOT of promise. And yes, it can do things that Midjourney cannot.
0:00 - Intro
2:25 - Firefly Text to Image UI
4:00 - Firefly Text Effects UI
5:23 - Prompt 1 Comparison
7:25 - Prompt 2 Comparison
8:20 - Prompt 3 Comparison
9:06 - Prompt 4 Comparison
10:07 - Prompt 5 Comparison
11:08 - Prompt 6 Comparison
12:05 - Prompt 7 Comparison
Let's get started!
#firefly #midjourney #ai
- - - - - - - - - - - - - - - - - - - - - -
Subscribe for NEW VIDEOS!
Learn UI/UX: designcourse.com
My personal FB account: logodesigner
Coursetro FB: coursetro
Coursetro's Twitter: / designcoursecom
Join my Discord! / discord
^-Chat with me and others
- - - - - - - - - - - - - - - - - - - - - -
Who is Gary Simon? Well, I'm a full stack developer with 2+ decades experience and I teach people how to design and code. I've created around 100+ courses for big brands like LinkedIn, Lynda.com, Pluralsight and Envato Network.
Now, I focus all of my time and energy on this channel and my website Designcourse.com.
Come to my discord server or add me on social media and say Hi! - Jak na to + styl
Very exciting times in AI. It just won't stop. Loving all of it, even if the singularity is reached shortly and we're all enslaved.
Excellent video.
6:16
In the MJ Office Hours yesterday, David did mention that the image data set for Midjourney was rather small in comparison with other AI image generators......which surprised me.
Aren't we already enslaved? lol
It's important for researchers, developers, and policymakers to work together to ensure that AI is developed in a safe and responsible manner, with ethical considerations at the forefront. This can help mitigate any potential negative consequences and ensure that AI is used for the betterment of humanity.
@@mixerjdp I really appreciate your comment
Enslaved?! Naa... It won't need any of us for anything. Except perhaps to give it some kind of purpose. For... you know... there is no purpose. So perhaps it might need us to provide it with some purpose.
Firefly is promising something better than midjourney and other AI for graphics. The ability to create vector graphic with AI and to make this editable is great.
holly shit thats just incredible, i wonder how those vectors are gonna look in the layers
My feeling as well re the vector graphic aspect. That, along with the ease of use promised by Adobe's interface with the myriad options makes it very appealing.
If Adobe makes it so there are layers in a photoshop format, they will be the go to given one could manipulate layers of the final product to fine tune results, add text, delete elements, etc.
My issue with Firefly at this point is a) you can not weight the prompt, b) you can not get a seed value, c) you can not include a link to an image to clarify your intentions. Both of them are just generative, but with Midjourney I can get more compositional control and build iteratively.
Props on hitting a million subscribers! Also, thanks for the product comparison...I was literally just wondering how Firefly stacked up against Midjourney yesterday.
Appreciate your efforts! Really helped it.
Thanks for the review... can't wait for my firefly invite
as it is now, based on the comparisons in this video, there is absolutely NO comparison between Firefly and Midjourney. With Midjourney V5 you can basically replace stock images, the photo-realism and quality are so good that in most cases you can not tell them apart from a real photo. Firefly is still way way behind. Of course this may change fast.
MJ's secret sauce is making some language model hallucinate really long textual descriptions first, then adding it to your prompt secretly. Firefly has the better architecture/potential in my view, and with some serious prompt engineering can probably TODAY produce near identical results to MJ5...I'm guessing...
Exactly. Not only are adobe images lower quality, they also have a stock-ish look to them. You might as well buy a real stock photo.
Awesome comparison of Firefly and Midjourney. Thank you so much.
I feel like firefly follows the prompt better, while midjourney has a better understanding of colors and composition. Overall, i still like midjourney more
Firefly images look like dull generic stock photos, which makes sense if itâs trained on stock photos. This is a big downside imo. No one wants to generate something that looks stock. You might as well just buy real stock photos.
Wow! really appreciated the detail and effort you put into this. I recently used Blue Willow to produce some incredible arts. If you could create a tutorial series about BW, that would be fantastic for the beginners like me
Great review / comparison, pretty impressive what firefly can do so far ! Looks promising
This was really really cool, do it more!
Thank you for going through the various examples you did and the comparisons. Firefly is not that far behind Midjourney at all. I do like the fact - being a creator of art - that Adobe is not stealing images off the web by scraping it like the other AI venues.
I would be hesitant to include such content from Midjourney etc. in a commercial piece I was producing for that reason. Copyright infringement lawsuits are very costly - both to pursue and defend - as someone who went after a magazine once for an infringement - and got them to settle out of court - thank goodness!
I would be interested in your opinion on this subject and what you envision as any possible complications resulting from the free for all acquisition of the variouos platforms of content without permission given for its use.
From my experience so far it is a big difference in V5 of MJ if you set stylize to around 300 and chaos to 25. MJ also stated that Stylize makes a big difference. Think it might have given even better results with these adjustment
You're right. All these tools require a lot of learning to use to their full (current) potential. Prompt engineering will make or break the job interview in the next few years. I'd show the person an image, and see if he knows what tool and prompt to use. I'd show him a result of a chat, and ask him to achieve the same result with his own prompt...
Iâm using Firefly on May 5th and I can see a marked improvement over what it was giving me back at the beginning of April.
Thank you - interesting comparison.
Looks like Firefly does a really good job of knowing the meaning of what youâre asking for, versus Midjourney which often ditches ideas it canât quite make fit.
Firefly just came out of nowhere for me, had no idea! I'm particularly interested in the text to vector tool, haven't been able to achieve that well with AI (Illustroke is okay, not amazing). Personally the Midjourney results looked better to me and Firefly puts up a good fight. I definitely appreciate the better UI and quickly changing styles, though! It's possible that Adobe just buys Midjourney and integrates their tech đ
Adobe can leap frog MJ easily...and they will, shortly... While I somewhat enjoy MJ, I think it's a bit overhyped. Vectorizing and allowing you to create models from your own content easily are way more important to me.
Itâs not over hyped at all. Just a year or two ago the sole idea of generating any image with just text was something out of a science fiction movie.
Very interesting. Thank you
Awesome!!, thank you so much for your time. So I am still confused which one is better either Midjourney or Firefly.
Awesome video. How do you zoom in so fluidly like that on Windows?
I have had the master collection subscription for tens years. Just found out about firefly. Signed up for invite. Hope I get in soon!
Ahhhhhgh just wish Firefly was better w text so we can start banging out logos sans Photoshop! Great vid
Great job
Woah, what app were you using to bring up 2 images side by side? That's slick!
default windows Photos. Just tick the checkbox from the list of images.
Question. Does anyone know what the resolution and file size limits that are projected for Firefly and also, if you know those details about Midjourney.
I am quite limited in terms of my use of the AI programs at this point - mostly DALL-E 2 use, which limits the size quite severely.
this is so exciting!
Leaving a comment to help the channel â€
Thank you for the comparison. For me, Firefly is going to be used most on the basis of ethical issues.
A question:
On the first prompt, in the MJ photo. Isn't there something off about the shadow of the woman in the chair? Or am I imagining things?
the mdijourney one, lady and a child, the picture you took as a good one.
i have a feeling the lady on the chair is missing her second hand
How did you show two images to compare? I mean the program and if is there a way to show it like that.
@DesignCourse what are the largest pixel dimensions available for Firefly-generated rasterized images?
I think this video is pretty solid.
I love the UI designs you post. But I am a question about Midjourney. Few days ago I added Midjourney in discord community to check myself how to get impressive UI designs. It is asking for subscription when I started writing in '/prompt' first time. Is it not free anymore.
Questions is will it be part of creative suite and still same price?
first yes, second no.. come on it's adobe, nothing's free
Thanks!
Out of the million AI products that have been launched, this is the only one I am truly excited about. Love MJ, but can't wait to play with this.. I am a professional artist who happily works AI into my designs.
"We're all probably going to be dead soon because it's going to take us over, but anyhow" đ
đđđ
Stable Diffusion is and will be always the best. The way you can control everything allows you to do anything on any resolution.
You shoukd get in touch with stable diffusion, it can do all of that.
Hi there, I just hit the button to request access and submit the filling, I wonder to know how long it takes?
Can they do basic stuff like forming charts, graphs on giving information? Say a chart on the Software development life cycle phases?
Now let's take FireFly and MidJourney, and blend them together. What do we get?
Text to vector is for Illustrator?
Personally, I like firefly's results more. Seems to be more professionally made pictures
I agree. MJ to my eyes look like a kid playing with filters on instagram...
I've overheard professional designers talking about AI art at work, and while nobody seemed to have dismissed it out of hand, it didn't sound like anybody was ready to take it seriously right now. Firefly might change that.
It's from a brand they're already deeply beholden to, and plus, it has features that I could actually see a designer using. (e.g. a more traditional style of UI, easy aspect ratios, typography tools, etc.)
You mentioned early in the video that Firefly was "ethically trained" on Adobe Stock? Do you have a source for that? I'm very interested in this point since, from a practical perspective, I know that for some people the way that these models are being trained can be a dealbreaker.
What I saw is a tweet from some Adobe official saying they trained on a combination of Adobe Stock and images in the public domain, or something like that.
Basically it's supposed to be free of any potential copyright issues.
@@user-wv1in4pz2w But aren't the allegations against the other major AI art models essentially that while they claim to have only trained on public domain datasets, that in actuality a lot of the stuff in those datasets was actually stuff posted to Pinterest etc. by people who were reposting original art without permission? Do we know that that's not the case here?
@@OverbiteGames These are complicated questions, so thank you for sharing your thinking.
I particularly agree with your point when it comes to "general" models trained on massive datasets. What right is being infringed when you can't draw a correlation from output, through the latent space representations in the model, back to a specific piece of training data?
I think it's a little more of a grey area when models are trained on art of specific artists with distinct styles though. Yes, that's a very hand-wavey way to frame the issue, but sometimes you see AI generated art or read about prompting techniques that are obviously and intentionally trying to rip an artist off. I feel like allowing that would be a pretty big and unintended disincentive.
Seems good (or at least as good as Midjourney)
I love MidJourney but it aggravates me more than not, so as an Adobe user, I am dying to get my hands on this. The possibilities are incredible.
Started today with Firefly. The images look some kind of vector art, or computer art but not like a real image taken with a camera, and the output is only a 1Mb jpeg image.
I tried these prompts with Bing image creator (Dall-E) and overall it sits between both. Sometimes worst, sometimes better than Firefly. Too bad we can't post photos on CZcams comment section.
Bing IC is based on DALL-E, not SD
@@urszulahkuzma Thanks for reminding me, I edited the post
@@elieh.studio you're welcome! đ
Leonardo is also a good one, it has few trained models too. you get 150 tokens per dayđ
This makes me want to buy the game Watchdogs. I think I will buy them both for my birthday next month.
It bothers me a bit that shadows never follow the real window structure. Just randomize lightning.
So basically just wait on firefly to improve. I like the options it offers
What about refining of your prompts? With MJ you can fix many bad details of first image by adding new keywords
I want to see how good it is at vectorising images. I mean, really well made vectors with economy and precision in the placing of nodes. Also, I'm not surprised yet - we already had cloning tools in PS, and 'fill content-aware' which acts very similarly to what's happening with typography, only that you are not deciding on the source: Looks like a text mask with a slightly more advanced content aware fill tool using unknown source material. It also relies heavily on the UI, which if it wasn't there, would make the software as annoying to use as Midjourney- couldn't exist without it. Another thing: in what way is this going to make a real difference for designers and instead just become as cliche and boring as 90s WordArt? So far AI images are absolutely great at generating textures and patterns, but I wonder if all this AI stuff that's coming out, with each software company presenting their apps as using AI - when sometimes it's just good old software -, apart from the intrinsic advancement in machine learning (all names that I think use rhetorical devices and marketing to anthropomorphize software), are not just more of the same things under a new packaging. I think it's called 'hype'.
What do you think about the fact that soon AI will make the website only from voice commands? Are our jobs at risk?
Midjourney most definitly has my votes.. Their coffee logos really got my attention compared to firefly.đ·
David Spade knows his shit!
For a beta, Firefly is looking great. The big question is: Can it ever catch up to Midjourney? Can Adobe's massive amount of cash bridge the time/learning gap that going first gives to Midjourney?
Of course they can,.within a few short months.... MJ isn't that impressive to me. You can get similar results by running your prompt first via a properly hallucinating (text) llm a la gpt4
Text management, generation and style is already an advancement compared to all other generative models atm.
You say "Midjourney can't do that" and it's true...but with v5's new tiling feature you could in theory tile a detailed texture image like that and then use background-clip: text; with a color:transparent; font to put a tiling texture image over text.
You could even combine shadow effects with a duplicated layer to get the 3d. But yeah a lot of work (and smallest MJ image you will get is still 512px which is not an ideal text tile)
And firefly will still be better because it morphs the texture to the text, and doesn't use precise cutouts for boundaries.
@@user-wv1in4pz2w yeah, definitely looks better. I'm still on the waitlist for an invite :(
AI just keeps surprising with every iteration. Recently I created amazing images with Blue Willow
Wow firefly is very impressive.. I think it's because Adobe collected a lot of picture and illustrations from Adobe suite users over so many years which makes such photo and vector data readily available for them to train their AI models.. Wow đź
It's all trained on public domain images and images from Adobe's stock library.
I am SUPER happy to see my job as a graphic designer is safe for now! Every thing Iâve seen sucks. WHOA! Any of you disagreeing, donât confuse illustration with graphic design.
I would love to be able to buy a product like Firefly and run it on my own hardware and produce things like my own icons, fonts, and generic artwork that would be able to be used without paying additional license fees.
Adobe should focus more on editable, piecewise art than full photos imo; If I need to add something to my photos: A background, an object, etc.
Fireflyâs text effects are great, and it is the single reason to use it. Their text to image generation is currently terrible - somewhere around MJ3. I hope MJ steps up to the challenge issued by Fireflyâs text effects, because that is frankly sick.
Thanks you for this early comparaison.
Can someone give me his access account for the Firefly beta :-) how mush time do i need to wait ?
All I know is that in the trial for mid journey is that you have to run down your creations :)
Adobe need some time to match Mid-journey quality
Honestly Firefly is comparable to Midjourney 3, but it has great potential and seems to understand promts better than Midjourney.
Firefly is a real creative tool combining AI feature. Others are pure generators.
You have a spelling error when it comes to the lamp in question ("on the corner" instead of "in the corner") and thus I would say it's more of a spelling correction comparison than output comparison (simply unfair). I'd like to see a written post with the results with correct spelling if there is any.
vector out of the box
so looking forward to the good video creation AI. Will that be in 2023?
It came out yesterday. Text to video.
@@elephantintheroom5678 I might be missing something here so please share which ones you are referring to?
Firefly is đ„
I think firefly will be better as it seems to produce images that are very close to the goal/target/intent. MJ creates a very cinematic looking image but maybe not exactly what the user intended.
Did the creators who uploaded their imagery to Adobe Stock know it would be used to train AI?
No clue, but I guarantee the terms allow them to do this. They are working out some sort of payment attribution model for them.
I have the feeling that midjourney has a mystical dark lighting by default
The type treatments are cool, but the uses are limited. I can easily see people overdoing it, until it becomes the new Comic Sans, where every design has textured lettering whether it needs it or not.
Canât wait for it to be integrated to photoshop, using my face to make wild thumbnails for YT
In a moment you can buy a car or just walk, you can get a PC or typewriter, use Photoshop or paint, yo get an Smartphone or a regular one
You started sounding a little.biased at the end, especially with the coffee logo generations. Mid journey clearly has the better logo generations in that instance but you said they look similar which isn't the case. Firefly no of the 4 from firefly looks like a logo at all.
Dude?! Are you related to Rob Halford? Never saw anyone who looked so much like the early Halford. Look up "Judas Priest - Dreamer Deceiver / Deceiver (BBC Performance)" on CZcams.
I would argue Midjourney clearly did better in every image, but the usability of Firefly is huge and I think they will catch up.
chetori ehsan?
Is this censored like mid journey? I canât imagine photoshop telling an artist what they are drawing is inappropriate but you never knowâŠ
I prefer Dall-E over midjourney. Don't know about firefly as I've never used it. midjourney has kind of a same theme for every image generation
It's crazy to me that the AI struggles with text. It's the one thing they should easily be able to get right because the user is inputting the actual text. The coders need to generate the AI art in the background, then add a layer over the top where they generate the text. Once that's working, they open the door to posters, business cards, logo's, stationary, etc.
yeah, i also was very surprised about that. Doesnt seem so hard from a humans standpoint đ
If you understand how the generation of images works, you'll understand the problems that come with it. Your suggestion would add self made logic to it, which derives from the pure neural network architecture but could be a specific application. However, the AI will become better at a fast pace and improve this as well and the specific application will be obsolete.
Deviates, not derives
@@Swoncey Yeah, I understand the tech and by all means keep training the models on text so that it can learn but in the meantime, there's no need for a purity approach when it comes to the end users.
MidJourney opened a giant lead on other art tools but all of us were disappointed at the text. Adobe's Firefly appears to have the text sorted, which will very quickly allow Adobe to reestablish their dominance when it comes to graphic design.
@@TheDandonian I get your point, but there is for sure possibility to add layers for text optimization and then the topic is solved. Why I'm saying this, is the following: Let's say you commanded the AI to print "abc" on a billboard. You cannot simply ignore the text in the image crearion process and as a postprocessing step add it to the billboard, since it is unknown where the billboard is located and whats the field of view etc.
Imagine just how easy it will be to make a movie with Ai-Generated videos
nothing beats midjourney
Although not perfectly, Firefly handles text way better
A lot of ai art remind me of clipart back in the day.
problem with hands and feets
Someday you will use Ai to create movies to watch and games to play. Always wished I could show people what was in my mind
With so many references for fonts, why is it that AI cannot handle simple typography but can paint an amazing face that is so much more intricate? From this video Adobe is not there yet. Adobe needs few more months to catch up with MidJourney
I wonder how good AI images will become in a couple of years. Right now most images look very creepy to me.
Just a reminder: 2 years ago, state of the art was Dall-e, not even Dall-e 2
I personally like Midjourney ones much MUCH better