Top 5 AI Image Generator Analysis (part 3)

Last Modified: Oct 9th, 2023 - Category: Artificial Intelligence, Case Studies, UX Research
AI Image Generators cover image
Best Generative Image AI Introduction
(part 1)
Defining the AI image generation apps to test
(part 2)
Top 5 AI Image Generator Analysis
(this page)
Experts View and Results
(part 4)

AI Image Generators: User research

Finally, the AI image generators page! I assume this part is the one most people will be interested in. It includes a thorough analysis of each tool through user research, something you won’t find anywhere else. But the last part of this case study includes the insight and results considering the experts’ views as well, so I’d recommend you pay a visit to that page (you can see the link above this paragraph).

Adobe Firefly

Adobe has been at the forefront of utilizing Artificial Intelligence in image processing. One of Adobe’s most significant contributions is the development of Photoshop’s Content Aware Fill, which marked a major milestone in the field. Given this extensive experience, it is unsurprising that Adobe has embraced the trend of generative image technology. In line with this, Adobe has introduced Adobe Firefly, currently in its beta phase.

UI for Adobe Firefly AI Image Generator

Adobe Firefly has 4 active modes in the beta version, with 8 additional modes in exploration.

In terms of the user interface, it’s no surprise that it looks clean and consistent, reminiscent of Photoshop, including the image filters section.

While everything in Adobe Firefly exudes sophistication and gives the impression of being an expensive product designed to satisfy even the most discerning users, the reality is that the results are below average when it comes to Text to Image (which is the focus of our research here). Most prompts I ran ended up with really weird results, including strange hands and feet (a common issue in generative AI), and even stranger faces, which is quite unexpected at this point. The photographic quality is average, and prompt adherence is slightly lower than what we could expect from Adobe.

Another unexpected aspect is the absence of a common feature: Image to Image generation. This feature is not even present as an “In Exploration” feature, which may suggest that Adobe Firefly will never offer this option.

Finally, once you create a set of images, if you try to download any of them, Firefly will add a prominently visible logo and the words “Not for Commercial Use” on top of a brightly colored background. This is quite interesting because they mention that they do this to be responsible about AI images. The topic of AI generation ethics is one we’ll cover soon.

User Research Results Sheet
Tester 1Tester 2Tester 3Tester 4Tester 5Average
Image quality433543.8
Prompt-to-Image332232.6
Image-to-ImageN/AN/AN/AN/AN/A0
TrainableN/AN/AN/AN/AN/A0
WeightingN/AN/AN/AN/AN/A0
Negative weightingN/AN/AN/AN/AN/A0
Special features453554.4
User Expectation423443.4
Usability343212.6
Absolute PriceN/AN/AN/AN/AN/A5
Relative Price111211.2
Total2.091
User research for Adobe Firefly AI Image Generator

Bing Free AI Image Generator

Bing Image Creator Interface

Bing is Microsoft’s search engine. You may wonder how it relates to generative AI.

Well, Bing has always struggled to gain a share of the search engine market, regardless of the tactics they’ve tried. So they needed to do something, and in my opinion, they found inspiration in You.com and simply copied their model. You.com is a search engine startup that aimed to be different from the very beginning. They offer the typical search process found on any search engine, but they also include chat-based search powered by AI, as well as image generation and other AI-based generative tools. Sound familiar?

Therefore, I don’t think it’s a coincidence. You.com predates Bing AI by at least a year, making it difficult to believe that it wasn’t the inspiration for Bing’s new AI features.

However, the reality is that Microsoft is a behemoth, and while they may not be able to compete with Google, they certainly can compete with everyone else. That’s why we analyzed Bing Image Generation, despite it having fewer models, simply because it’s more popular when it comes to AI generated images.

Bing AI Image Generation is completely free and operates on a credits system. Users are allocated a certain amount of credits per day, which can vary. Additionally, you can also earn free credits by using the tool. The default amount of tokens is 100. The image generation engine is a custom Dall-E model.

In terms of interface and user flow, it’s a bit confusing. Unless you know exactly what you’re looking for, it’s quite difficult for novice users to find out how to create images.

The flow goes as follows: Go to the Bing homepage, click on the images tab, and then on the top right of your screen, you’ll find the “Create Images” button. Although it stands out with a bright color, it’s positioned so far away from the search box that it can be easily missed on large screens like mine.

Once you reach the image generation page, users will see a prompt box and two buttons on the right side. The first button is used to submit the prompt, while the second one (labeled as “Surprise me!“) generates random prompts.

Bing Free AI Image Generator doesn’t provide any options. It generates four variants of the prompt as output. These images are sized at 1024×1024. When you click on any of the images, you’ll see four options: Share (the main call to action or CTA), Save, Download, and Customize.

The last option is the most interesting to us because it could be the place to fine-tune the output. However, the link leads only to Microsoft Designer presets and templates that use the AI-generated image.

In short, it offers just a prompt, no options, and four 1024×1024 images. It’s fun and free to use.

User Research Results Sheet
Tester 1Tester 2Tester 3Tester 4Tester 5Average
Image quality221442.6
Prompt-to-Image433443.6
Image-to-ImageN/AN/AN/AN/AN/A0
TrainableN/AN/AN/AN/AN/A0
WeightingN/AN/AN/AN/AN/A0
Negative weightingN/AN/AN/AN/AN/A0
Special features111211.2
User Expectation334433.4
Usability333433.2
Absolute PriceN/AN/AN/AN/AN/A5
Relative Price111211.2
Total1.836
User research for Bing Free AI Image Generator

Leonardo.AI

Leonardo is a very intriguing tool within the world of generative image AI. Originally aimed to create game assets, it evolved to a massive tool with lots of features. On top of that, it uses a lot of models, including proprietary models, Dall-E, Stable Diffusion and (get ready)… user generated models!

Generative Image AI: Leonardo.AI prompt commands
Generative Image AI: Leonardo.AI prompt commands

Another remarkable feature of Leonardo is that it’s completely free at the moment. Users are provided with 150 tokens per day, which they can utilize for various features such as image creation, fine-tuning, and resizing. Each operation has its associated token cost. For example, creating an average image with four options costs 4 tokens, while upscaling an image costs 5 tokens. If users exhaust their token allocation, they have the option to subscribe to one of the plans starting at $10 per month, which grants them 8500 tokens.

Leonardo.ai operates with a waiting list, meaning that you can’t simply sign up on their website and gain immediate access. On the contrary, you will need to wait for a certain period of time. However, you can expedite the process by joining their Discord community and completing a form with some personal information. I went through this process myself, and it took approximately a week to get my account approved.

Upon entering Leonardo.ai, you may find yourself overwhelmed by the multitude of possibilities it offers. The homepage can be a bit daunting, but if your objective is image generation, you should navigate to the “AI Image Generation” tab located in the sidebar.

Clicking on that tab will redirect you to the image creation page. Similar to the homepage, this page features a sidebar on the left and a main content area on the right. The sidebar contains most of the fine-tuning options, while the main content area includes the prompt window, model selection, and the generated art once you start creating images.

Overall, considering the complexity of the tool and the abundance of features, the UI design is exceptionally well-crafted. However, it could be improved in terms of cognitive and visual accessibility. Aside of this, it functions well from a UX standpoint.

Going deeper into Leonardo.ai, the most noteworthy aspect is the wide variety of models available. These range from Stable Diffusion to proprietary models developed by Leonardo, as well as featured models trained by users. Additionally, there is a feature called “Alchemy,” which is particularly useful for creating eye-catching realistic AI generated images. Post-generation features include upscaling with different modalities, background removal, and a one-step zoom out capability (meaning you can only zoom once).

But that’s not all! Leonardo.ai also offers two additional modes in addition to Generative Image AI: 3D Textures (which allows you to upload a .obj file and render it with textures) and AI Canvas (which enables the creation of complex scenarios by combining different AI-generated images). The former mode is specifically tailored for game developers and is currently in the Alpha stage, while the latter is in Beta.

In short: a very high quality free AI image generator. Is it possible that there’s a better tool than this? Let’s see!

User Research Results Sheet
Tester 1Tester 2Tester 3Tester 4Tester 5Average
Image quality543444
Prompt-to-Image445444.2
Image-to-Image323343
Trainable555555
Weighting314222.4
Negative weighting214453.2
Special features555454.8
User Expectation534354
Usability214322.4
Absolute PriceN/AN/AN/AN/AN/A5
Relative Price545554.8
Total3.891
User research data sheet for Generative Image AI using Leonardo.ai

Kaiber

Kaiber sets itself apart from other generative image AI tools by offering a unique feature — it creates videos instead of static ai generated images. This innovative approach adds an exciting dimension to the creative possibilities. Priced at $15 per month, or $12 per month for annual subscriptions, Kaiber provides an accessible and affordable platform for video creation.

In addition to video generation, Kaiber offers several impressive features. Users can explore models designed by renowned artists, experiment with camera orientation, and even generate videos from existing video clips, text prompts, images, or audio files. Combining these elements allows users to craft dynamic and multi-layered videos. For example, users can upload an image, add a text prompt, and incorporate an audio file in order to create a video.

It’s worth noting that the proportions of the uploaded image or video heavily influence the final video output. For instance, if you upload a 2:3 image but desire a 16:9 video, the system won’t accommodate this. Instead, users must either upload a 16:9 image or opt to exclude any images or videos entirely. Kaiber compensates for this limitation by offering users proportion options if you don’t upload any image or video, a discovery I made by chance. Additionally, you can use another Ai image generator to create the images as you want, and then upload it to Kaiber.

Once content is uploaded and prompts are generated, users proceed to a second step where they can customize camera movements, determine video duration, specify the level of creativity, and even adjust reactivity to music. The latter feature enhances the video’s synchronization with the rhythm, making it particularly appealing for rhythmic songs.

Like other generative image systems, Kaiber operates on a token-based system, charging 1 token per second of video generation. Consequently, a 10-second video would cost 10 tokens. The generation time varies depending on the uploaded content and video length. However, Kaiber’s processing time is relatively low compared to traditional video rendering methods. A 30-second video can be generated in less than 10 minutes, a noteworthy improvement for video creators accustomed to lengthy rendering periods.

After generating a video, users have the option to upscale the output to either 1080p or 4K resolution, adding a touch of refinement to their creations.

Now, despite the many positives of Kaiber, there are some drawbacks that warrant consideration.

The user interface (UI) design is generally satisfactory. However, some users may find it slightly confusing to distinguish between functional buttons and decorative elements. From a visual accessibility standpoint, the UI offers acceptable contrast. However, it falls short in terms of compatibility with ARIA/WAI and assistive technologies. While it’s understandable that visually impaired users may not gravitate toward a visually-oriented tool, Kaiber’s cognitive accessibility leaves much to be desired.

The most important drawback: Kaiber is an AI “video-ish” tool. Its video generation capabilities are limited to morphing 3D-generated images. It cannot create real video. For example, it can’t create people walking, but morph images, as evident in an example prompt:

figures walking, one of them is a female, they appear one by one, the female is the last one, they walk through a desert, fast motion

in the style of film photograph, black and white, film grain, highly detailed, masterpiece, realistic, soft lights, cinematic focus

Kaiber prompt

Weird to say the least.

Also, you can see the Kaiber watermark in the video. This video was created with a paid account with a hidden automatic renewal (which was reason enough to cancel the service). But despite still having credits, they work on a 30-day subscription model on top of the tokens, whichever happens first.

So, even though I have remaining tokens, I’m not allowed to create new videos, and I can’t download non-watermarked videos I have created (and paid for), with one exception: if I had upscaled them (which has an additional cost), then those are available without a watermark.

Top 5 AI Image Generator Analysis (part 3) 1

Finally, let’s talk about the support provided by Kaiber, which is through Discord. There was a day where nothing was working at all, and I mean absolutely nothing. So I had to resort to reaching out for help on Discord. However, when I presented the evidence of the issue to Kaiber’s representatives, they denied it, despite other users also complaining about the same problem. Later that day, they finally fixed it. But the user experience of being lied to by customer support is extremely disappointing and negatively impacts the overall customer experience (CX), adding up to the previous issues mentioned.

In short: Kaiber offers a platform for creating fun and imaginative videos, particularly suitable for casual use and sharing on social media. However, the platform’s business practices and ethical considerations are questionable, impacting the overall user experience. Kaiber has the potential to be a game-changer in the field of video generation if improvements are made, but its long-term success remains uncertain.

User Research Results Sheet
Tester 1Tester 2Tester 3Tester 4Tester 5Average
Image quality453454.2
Prompt-to-Image132422.4
Image-to-Image343453.8
TrainableN/AN/AN/AN/AN/A0
Weighting233312.4
Negative weighting131211.6
Special features453243.6
User Expectation121221.6
Usability123312
Absolute Price234455
Relative Price332132.4
Total2.636
User research sheet for Kaiber AI Video generation

Midjourney

Our final generative image AI tool is Midjourney.

Midjourney is truly exceptional in every sense: it was self-funded, and it is 100% owned by the team. It operates through prompts on Discord and utilizes an exclusive, undisclosed model. In short, it is an incredibly intriguing artificial intelligence tool.

As mentioned earlier, Midjourney operates entirely through Discord. This means that it inherits all the inherent issues associated with Discord (which are significant) while also introducing its own unique challenges.

However, despite these factors, Midjourney is relatively easy to use once you become familiar with it. Instead of having its own dedicated user interface with clickable buttons or dropdown menus for selections, it primarily relies on text prompts. The only exceptions are the “image2image” feature, which generates images from existing ones, and the blending functionality that allows users to upload up to four images and blend them into a single image.

Perhaps due to its text-based approach, Midjourney stands out between AI image generators because of its exceptional prompt quality. No other tool even comes close to matching it. The predefined commands are easy to grasp, starting with the basic “/imagine” command, and the weightings are remarkably accurate. Users have the flexibility to define almost any aspect of the desired output, including resolution, quality, prompt adherence, model selection (limited to two versions), infinite zoom, infinite pan (in a single direction), and more.

AI Generated Image Tools: low chaos example
Generative image AI low chaos example

All parameters are well-documented, with examples provided for each parameter to demonstrate their functionality and the modifications they can make to the output.

Midjourney’s stands out, lo and behold, as the AI image generation tool with superior image quality and excellent prompt adherence. Despite its costliness compared to other tools (priced at $8 per month for 200 images, billed annually, or $10 monthly), it remains the most popular choice. And it’s no surprise; every feature works exactly as expected and even surpasses expectations.

In short, an astounding AI image generator with remarkable results, suitable for both recreational and professional use.

On the flip side, it poses some challenges for novice users as it requires familiarity with two apps (Discord and Midjourney). There is a learning curve, and it’s relatively expensive (compared to other tools). Additionally, similar to Kaiber, it employs a hidden renewal trick. Users won’t realize they will be charged again unless they explicitly instruct the credit card processor (not Midjourney!) to stop the charges.

User Research Results Sheet
Tester 1Tester 2Tester 3Tester 4Tester 5Average
Image quality555555
Prompt-to-Image534544.2
Image-to-Image443554.2
TrainableN/AN/AN/AN/AN/A0
Weighting445243.8
Negative weighting443323.2
Special features555544.8
User Expectation555555
Usability132111.6
Absolute Price434235
Relative Price555344.4
Total3.745
User research sheet for Midjourney AI image generation

AI Generated Image Tools: Continuation

This is the third installment of a four-part case study on tools for AI generated images. If you’re interested in reading the expert’s perspective and final insights, please continue to Generative AI Tools: Experts View and Results. Alternatively, if you arrived from a different website, kindly scroll to the top of the page to access all sections of this study.

We can improve your business!

Let us help you with the best solutions for your business.

It only takes one step, you're one click away from getting guaranteed results!

I want to improve my business NOW!