AI Artistry (Part II): Which is the All Round Best Image Generation Platform?
We compare the full value proposition of six leading AI image generators including Midjourney, Adobe Firefly, Leonardo AI, Microsoft Designer, Ideogram and Flux
Following our in-depth comparison of image outputs from six leading AI image generation platforms, this article takes a broader perspective. Beyond image quality, we'll today assess factors like ease of use, pricing, licensing, customization options, copyright considerations, processing speed, and project support, with the aim to guide non-technical users in selecting the best tool for their creative needs.
Assessment Criteria
This assessment is based on my own testing and evaluation, supplemented by company announcements and other analyst reviews. I have focused on usability and relevance for non-technical users, and have rated each platform along the following dimensions on a scale of 1 to 5 (with 1 being worst and 5 being the best):
Image Quality: Clarity, realism, resolution, overall visual appeal of generated images, and prompt adherence (which is the degree to which the image generated aligns with what has been requested in the prompt). Image quality was analysed in detail in the first article of this series.
Ease of Use: Intuitiveness and navigability of the user interface, the learning curve for new users, and the availability of guidance and support resources.
Pricing: Pricing model and overall cost/
Licensing: Terms for using and distributing generated images.
Customisation: Range and granularity of adjustable parameters to support specific user needs.
Copyright Considerations: Service provider’s policy towards and handling of potential copyright issues.
Processing Speed: Time taken to generate images and secondarily, the capability to handle batch processing (i.e., multiple image generations) efficiently.
Project Support: Ability of the tool to integrate with other creative software and support complex creative workflows.
#1 Midjourney
Midjourney is the leading image generation platform, and the go-to tool for users that prize image quality, stylistic variety, and the highest degree of customisation over outputs. Despite its powerful capabilities, Midjourney's reliance on parameter commands (see below) and the lack of a free tier may deter casual users.
Here is how the platform fares across various dimensions:
Image Quality (5/5): Midjourney produces very high quality images that are visually striking and highly detailed. It shines in its ability to produce a wide variety of artistic styles and with its high degree of prompt adherence. The latest model, Midjourney v6.1, largely eliminates the common errors (e.g., inaccurate rending of teeth, fingers) that plagued the early days of AI image generation.
Ease of Use (3/5): Midjourney has until recently only been available on Discord, an instant messaging platform, which can be less convenient for those not used to the tool. However, the newly released web application, Midjourney Alpha significantly improves usability. The need to learn parameter commands (see below) has meant that Midjourney has a slightly steeper learning curve than other tools. On the other hand, the highly active community of 17 million users (as of end 2023) and the broad social media following have produced libraries of guides and courses so that new users need not need look far for support and inspiration.
Customisation (5/5): Midjourney’s long list of adjustable parameters allow for near infinite variation and the highest degree of output customisation among image generation platforms. Examples include aspect ratio (to set image dimension), style and character references (to use the same style and characters across images), chaos (to toggle the degree of creativity) and many more.
Pricing (3/5): There is no free tier. Plans range in price from US$10 (Basic) to US$120 (Mega) / month. The basic tier offers 3.3 hours of generation time / month, equivalent to roughly 288 image prompts. Most day-to-day users will find the Basic plan to be more than sufficient for their needs.
Licensing (3/5): Users are “free to use [their] images in just about any way [they] want”. However, if users work for a company with >US$1m in revenues, they need to subscribe to the Pro (US$60 / month) or Mega plans to have the same rights.
Copyright Considerations (1/5): Midjourney has not disclosed its training data but it likely involves the use of public artworks and other copyrighted sources. The terms of service publicly state they do not bear any liability for potential copyright infringements while using the platform.
Processing Speed (3/5): The processing speed for each image prompt is now 45 sec, which is marginally slower than most other platforms. Midjourney’s “repeat” parameter command allows users to quickly run the same prompt multiple times, and its “permutation prompt” can be employed to rapidly run variations of the same prompt (e.g., red bird, blue bird, yellow bird).
Project Support (2/5): Midjourney is a standalone image generator. While the platform has basic image editing (e.g., inpainting and outpainting) capabilities, users need to export images to other platforms such as Adobe or Canva for complex editing or if used as part of broader workflows. Recently released new image editing capabilities for its web application have added a few more features.
#2 Adobe Firefly
Adobe Firefly integrates seamlessly into the Adobe Creative Cloud suite, making it ideal for those already invested in Adobe's ecosystem. While its image quality is less consistent compared to competitors, its strong copyright indemnity for paid users can be highly valuable, particularly for enterprise users concerned with legal protections.
Here is how the platform fares across various dimensions:
Image Quality (2/5): Images created by Firefly are relatively hit-and-miss and have noticeably limited artistic variety. Firefly also struggles with detailed human features and is less likely to recognise copyrighted concepts (e.g., a Stormtrooper from Star Wars, Disney’s Mickey Mouse) because its training dataset is mostly limited to Adobe’s library of images (more on this below).
Ease of Use (5/5): Firefly offers a simple and intuitive interface that is easy to navigate. Rather than needing users to specify artistic elements such as aspect ratio, style, composition, or lighting in the prompt, the interface offers a modest menu of options. For instance “Bokeh Effect”, “Layered Paper”, “Digital Art”, and “Hyper Realistic” are some of the pre-designated choices on offer. “Suggestions” is an inbuilt feature helpful for those new to image prompting, automatically offering suggestions on how to expand and improve one’s prompt.
Customisation (3/5): Firefly’s menu of options approach should be sufficient for those less familiar with the nuances of artistic elements, or those looking to quickly generate decent quality images. However it is unlikely to satisfy users seeking professional standard images or highly stylistic outputs.
Pricing (4/5): Firefly can be accessed as part of Adobe Express, which is a creative content platform for creating videos, PDF documents, graphics and other digital assets. The free plan offers access to all Firefly feature but with only 25 Generative AI credits / month, which is enough for only a few prompts. The paid plans start at US$10 / month, comparing favourably with standalone platforms such as Midjourney and Ideogram, which are similarly priced but lack broader content creation functionality. Subscribers to Adobe Creative Cloud (US$60 / month), the umbrella suite for all of Adobe’s tools, also have access to Firefly.
Licensing (5/5): Adobe has explicitly stated that Firefly-generated outputs can be used for commercial projects.
Copyright Considerations (5/5): Adobe claims that most of its training dataset has been derived from its own library of images, for which it owns the copyrights. This allows the company to offer the aforementioned indemnity, and is one of Firefly’s key selling points to enterprise customers. If your goal is to use images in a situation where there could be sensitivity around copyrights, then Firefly (and other platforms that offer similar indemnities) could rank near the top.
Processing Speed (4/5): Firefly is very quick, typically taking between 15-20 sec for most image generations. However, batch processing is only available through APIs and not the standard user interface.
Project Support (5/5): Its integration into Adobe Express and Creative Cloud is a key selling point for Firefly, allowing users to create, edit, and deploy generated images as part of existing creative projects within these platforms.
#3 Leonardo AI
Leonardo AI is favored, particularly among game developers and concept artists, for its high degree of customisation while maintaining ease of use. Although it trails behind Midjourney in artistic variety, its ability to train specific models with user images sets it apart. As the platform integrates with Canva, which recently acquired the company, its potential for broader content creation is expected to grow
Here is how the platform fares across various dimensions:
Image Quality (4/5): Leonardo AI creates high quality images, which I would rate as fourth (behind Midjourney, Flux, and Ideogram) in this list. It does well with photographs and concept art, but lags in artistic variety and in having less visually striking and detailed images. Leonardo AI has mostly addressed issues with inaccurate rendering of human features, but these do continue to pop up from time-to-time, especially when images involve multiple people.
Ease of Use (5/5): Leonardo AI is as easy to use as Firefly, offering the same menu of options approach, with choices for a range of styles (e.g., Cinematic, Graphic Design Vector, HDR), aspect ratios, image size etc. The platform’s 4 million strong user community has produced a wide range of how-to guides, although not to the same depth and breath as Midjourney. “Prompt Enhance”, like Firefly’s Suggestions feature, automatically enhances users’ prompts to improve detail.
Customisation (5/5): While Leonardo AI does not offer the near infinite degree of control that Midjourney does, it makes up for it by allowing users to easily train their own model. As a marketer that wants to create images aligned with a brand guidelines, or a game developer looking to preserve the same aesthetic across all concept artwork, this capability can be extremely powerful. Users less inclined to train their own models can select from ten pre-trained models, each specialising in different types of images (e.g., anime, stock photography, portraits).
Pricing (3/5): The free tier offers 150 image credits (~10-20 image prompts) per day but with a very limited feature set. The paid plans range from US$12 (Apprentice) to US$60 (Maestro Unlimited) / month. As with Midjourney, the lowest paid tier (Apprentice) will likely satisfy the needs of ordinary users.
Licensing (4/5): Paid subscribers may use generated images as they wish. Leonardo AI owns the rights to images generated under the free tier.
Copyright Considerations (2/5): Leonardo AI’s current terms of service preclude any user protection for infringement on third party rights, including copyrighted data used in its training dataset. However, the company is reportedly in the process of establishing a “copyright shield”. The company’s acquisition by Canva, a significantly more established business, should hasten this development.
Processing Speed (4/5): Images are generated within ~30 sec which is very quick. While Leonardo AI offers APIs to developers that support batch processing, the website application only supports generating a small number of variations of the same prompt (four variations for the free tier and eight variations for paid users).
Project Support (4/5): Leonardo AI, while currently a standalone image generation tool, offers fairly advanced image editing and composition features. These include Canvas Editor, which supports moderate complexity edits to images, just like a “lite” version of Adobe’s Photoshop). The Realtime Canvas feature allows users to sketch out ideas and immediately see them transformed into realistic images. Transparency Mode enables the creation of images with transparent backgrounds. The ability to train one’s own models for specific projects is a huge plus. Longer term, the platform’s acquisition by Canva will likely also lead to some integration with Canva’s content creation platform.
#4 Microsoft Designer
Microsoft Designer, the only fully free tool among these platforms, combines Dall-E 3's image generation with basic content creation features. While it doesn't match the quality of top competitors, its simplicity, versatility, and integration with Microsoft 365 make it an accessible starting point for beginners
Here is how the platform fares across various dimensions:
Image Quality (3/5): Designer clearly ranks behind Midjourney, Flux, Leonardo AI, and Ideogram in terms of quality, including realism and level of detail. When facing off against Firefly however, Designer has the upper hand when it comes to artistic creations and variety of styles, whereas Firefly appears to produce better photographic and photorealistic outputs.
Ease of Use (5/5): Designer is extremely easy to use, with nothing more than a chat box for prompt inputs. Users need only select from one of three choices for the aspect ratio (i.e., square, portrait, wide). For brand new beginners to AI and image generation, Designer is a cheap and cheerful place to start exploration.
Customisation (2/5): Designer offers only the single option to toggle the aspect ratio, and therefore has the lowest degree of customisation and control of the tools examined. It does however offer a modest set of “smartphone-level” capabilities for image editing, which can be helpful for minor modifications (e.g., adding text and filters, adjusting brightness and contrast, removing background).
Pricing (5/5): Designer is entirely free! Yay!
Licensing (2/5): The terms of use, as of December 2023, explicitly states that it is “for personal use only and not for use in the course of trade or commerce”.
Copyright Considerations (2/5): Microsoft has a copyright shield that indemnifies customers of its “Copilots” in the event of copyright claims, but coverage appears to only include paid services, which would exclude Designer if used as a free tool. Designer can however be used as part of the Microsoft 365 Copilot package, so there is a chance that potential copyright issues could be covered in such case.
Processing Speed (4/5): Images are created within ~15-20 sec, putting Designer on par with Firefly as the fastest platforms on this list. However, there is no batch processing capability available.
Project Support (5/5): While not at the same level of functionality as Adobe Express and Canva, Designer offers versatile content creation suitable for simple presentations, quick social media graphics, event invitations etc. It is capable of creating not just the images, but the entire content from a simple prompt. For instance, “Create an invitation for Johnny's 7th birthday on 1st Sep 2024 at Green Park. RSVP at +44 (0) 1234 567 890. Make it aquatic themed”. Users can then easily edit the text and other elements of the graphic. Designer’s direct integration into Microsoft 365 apps such as Word, PowerPoint etc. is another benefit.
#5 Ideogram
Ideogram’s latest 2.0 model makes it a strong contender with Flux and Midjourney for the top spot. Where it stands out is in its ability to produce accurate text-in-images. This makes it invaluable for creating posters, logos, and other text-heavy visuals.
Here is how the platform fares across various dimensions:
Image Quality (5/5): Ideogram sits closely behind Midjourney in terms of image quality, with quality outputs and very good prompt adherence. The newest Ideogram 2.0 model significantly improved accuracy of human features and artistic range. Ideogram excels in text-in-images (something which all other platforms struggle with) and usually requires no more than a maximum of 2-3 reruns of the prompt to get this right.
Ease of Use (4/5): As with Firefly, Leonardo AI, and Designer, Ideogram has adopted an easy to use interface based around the menu of options approach. However Ideogram ranks slightly below on Firefly and Leonardo AI as it lacks an automated prompts completion / prompts enhancer feature.
Customisation (2/5): Ideogram’s menu of image customisation options is limited, primarily offering only aspect ratios, the ability to create tiled images, and the use of image seeds. The platform is a newcomers to the market, so I expect the level of customisation offered to improve over time.
Pricing (4/5): Ideogram offers a free tier which allows up to 10 “slow” image prompts (see below) per day. The paid plans range from US$8-60 / month and offer both “slow” and “priority” prompts (more on this below).
Licensing (5/5): According to the platform’s terms of service, Ideogram does not claim any ownership rights with respect to user inputs or generated outputs, and the company does not restrict users’ ability to employ the generated outputs for their own purposes, including for commercial purposes.
Copyright Considerations (1/5): Users “are responsible for [their own content], including taking all steps necessary to ensure that it does not violate any laws or rights of third parties”. No specific copyright protection is mentioned.
Processing Speed (3/5): Images are created within ~20-30 sec. Slow prompts are dependent on server capacity, with users potentially needing to wait up to 180 sec before image generation starts. Priority prompts are not limited in this way. There is currently no batch processing capability, although Ideogram’s roadmap includes a feature called “Relax bulk generation with CSV” for the Pro plan, which will presumably allow users to upload a CSV with a list of prompts.
Project Support (3/5): Ideogram offers image editing tools and features which are marginally more advanced than Midjourney’s. With Ideogram Editor (available only to paid users), images can cropped, resized, and adjusted with an inpainting feature. Text and the type of lettering used can also be easily added or modified.
#6 Flux
Flux’s release in late July 2024 received rave reviews. Unlike other entries on this list, Flux is a model rather than a standalone product per se and comes in three variants.
These are the Pro, Dev, and Schnell variants. Pro produces the highest quality images, while Schnell is the fastest. Dev and Schnell are open-weight and open source respectively, meaning that anyone can download, use, and modify the model (provided you have the technical know-how). Less technical users will for now need to access Flux via third party platforms such as Poe, Replicate, BasedLabs, NightCafe Studio etc.
Here is how the platform fares across various dimensions:
N.B. Beyond image quality, my assessment does not provide a specific rating for Flux across other dimensions because these factors are dependent on how the model is being accessed (e.g., downloaded and run locally or via a third party platform).
Image Quality (5/5): Industry observers have found Flux to be either on par with or nearly as good as Midjourney in terms of image output, which I have corroborated through my own testing of the highest quality model, Flux Pro. Flux excels in its ability to precisely handle complex prompts, and with high-quality photographs and photorealistic images. Midjourney comes out ahead in terms of its ability to get creative and produce a wide variety of styles.
Ease of Use: For users that want to download, use, and potentially modify the model, there are several technical requirements, including the non-trivial need for sufficient computing power to run a 12 billion parameter model. For non-technical users, the ease of use and user experience depends on the third party platform selected.
Customisation: Users that download and run the model locally have a high degree of control over outputs, and those savvy enough to work with the open-weights models can technically even train versions of the model for their own specific use. However, accessing Flux via third party platforms will typically only allow basic control over factors such as aspect ratio, tiling, contrasts etc.
Pricing: Dev and Schnell are available free-to-use if run locally on a user’s computer. Most third party platforms that offer Flux provide the first few images free for testing purposes, then charge for access thereafter.
Licensing: Pro is a proprietary model and its outputs are available for commercial use. Dev only allows for non-commercial usage. Schnell is available under an open source Apache 2.0 license which means that users may use, modify, distribute and incorporate the model in proprietary software including for commercial use, but must provide proper attribution. When accessed via a third party platform, the licensing terms of Flux may also interact with the former’s terms of service, which users should make sure to check.
Copyright Considerations: Black Forest Labs has said nothing about potential copyright infringements, and does not provide users with any sort of copyright protection against such claims. Its models seem to have less much less safeguards than other models, which could lead to potential copyright and other legal issues further down the line. For instance, users have been able to create images such as deepfakes of “Donald Trump flying a plane into the World Trade Center buildings and the prophet Muhammad holding a bomb, as well as depictions of Taylor Swift, Kamala Harris and Alexandria Ocasio-Cortez in lingerie”.
Processing Speed: The answer is “it depends”. While we know that Schnell is the fastest of the three models and that Pro and Dev are also very quick, actual processing speed depends heavily on computing speed and capacity, whether you are running the models locally or accessing it through a third party platform.
Project Support: There is no platform project support for users running Flux models locally. For now, most of the third party platforms offering access to Flix offer no or limited project support capabilities.
Conclusion
As the landscape of AI image generation continues to evolve, the choice of platform depends heavily on your specific needs and creative goals. There is really no one size-fits-all option, at least for now. While Midjourney stands atop the rest with its artistic output and depth of customisation, Flux and Ideogram are clearly not far behind. And While Adobe Firefly is less consistent in the quality of its outputs, it does offer peace of mind with its robust copyright indemnity. With what you now know about image generation tools, the ultimate question is—what will you create next?
Justin Tan is passionate about supporting organisations and teams to navigate disruptive change and towards sustainable and robust growth. He founded Evolutio Consulting in 2021 to help senior leaders to upskill and accelerate adoption of AI within their organisation through AI literacy and proficiency training, and also works with his clients to design and build bespoke AI solutions that drive growth and productivity for their businesses. If you're pondering how to harness these technologies in your business, or simply fancy a chat about the latest developments in AI, why not reach out?