Redefining Text-to-Image: How BestBanner Outperforms Midjourney and Stable Diffusion - A Benchmark Analysis
BestBanner is an AI-powered text-to-image generation system. It takes article text as input and generates high-quality social media banners. This saves you from the task of crafting images and lets you focus on creating content.
Another week, another product announcement! Our engineers have been working day in and day out, and burning the midnight oil to bring you Jina AI’s latest tool in the generative AI toolbox: BestBanner.
BestBanner is a unique AI-powered text-to-image generation system. It takes the text from an article as input and generates high-quality social media banners. This saves you from the time-consuming task of crafting images and allows you to focus on creating great content.
It’s as simple as pasting your article (or tweet) text into BestBanner and hitting “Go”.
No need to worry about how long your post is, and whether the model can handle that much information. And no need to deal with janky Discord UIs to generate your images either.
Wait a few minutes and you’ll get four output images, each of which you can choose to refine.
Next up, click the refine button to get high-quality variants of your desired style:
BestBanner will now take that image and riff on it, giving you four variations:
After a round or two of refinement you’ll have the banner image of your dreams:
Once you’ve found something you like, you can preview how it’ll look on your social network of choice, crop accordingly, and download for use:
Whether you’re a writer, content creator, or tweeter/skeeter/tooter/whatever-the-new-thing is, BestBanner is a great tool to add to your creative arsenal.
Why are social media banners important anyway?
The power of compelling social media banners for articles, blog posts, and other content is critical in today's digital landscape. As a content creator, your banners are essentially the digital storefront of your work, your first opportunity to capture the attention of potential readers amidst the digital clutter of their feeds. An eye-catching banner doesn’t just instantly communicate the essence of your content, but also entices viewers to click through and engage with your material. In a highly visual online world, these banners can dramatically elevate the visibility and impact of your content. What's more, many social media platforms' algorithms prioritize visually rich and engaging content, meaning your carefully-crafted banners can also boost your reach organically. Therefore, investing in the creation of compelling social media banners is a vital aspect of a successful content strategy.
BestBanner is an indispensable tool for making these banners happen, given its innovative application of AI-powered image generation. By allowing you to effortlessly convert blog posts, tweets, and articles into compelling social media banners, BestBanner eliminates the time-consuming task of manual graphic design. You simply copy and paste your content, and with a click, generate a selection of candidate images finely tuned to the essence of your post. These images can be further refined and downloaded, providing you with a visually striking banner that resonates with your audience and enhances engagement.
How is BestBanner different from Midjourney, Stable Diffusion, et al?
Don’t get us wrong - we love Midjourney and Stable Diffusion. They’ve completely changed the game in generative AI. But BestBanner differs in several key ways:
Copy and paste, don’t think: No need for you to wrack your brains thinking of what would make a perfect image for your article. Just paste the whole thing in and smack go. You don’t even need to remove formatting or surrounding HTML.
Hyperfocus: Midjourney, Stable Diffusion, DALL-E 2 and the others are all general-purpose image generation models. So they’re a good tool whether you want to generate a social media avatar, a logo or a poster, they’ll all do a reasonable job. But since they don’t focus on one thing in particular, they become Jacks of all Trades, and masters of none. On the other hand, BetterBanner is specialized for social media banner creation.
Unlimited context length: Other models are all designed to use prompts — short snippets of text that require a lot of tuning to get good results. BestBanner goes a step further and lets you use the whole text of your blog post, article, or tweet. You don’t need to think about what the most relevant point of your wall of text is - BestBanner works out what’s important.
Multi-model: While Midjourney and Stable Diffusion let you generate images with one model at a time, BestBanner uses three models to generate four images. This gives you a more varied array of stylistic and subject choices to choose from when it comes to refinement
Easy API access: Midjourney is only usable through their Discord bot, while BestBanner has an accessible API for you to programmatically generate banners with ease.
No censorship: In our testing, both Midjourney and Stable Diffusion often censor on the prompt level, immediately barring us from creating certain images. In other cases, Stable Diffusion simply generates sterile generic content given certain prompts. (None of these prompts were in any way obscene - often they referenced politics or were entirely innocuous). BestBanner has shown no such political or random censorship.
Showdown: Battle of the Banners
So, how does BestBanner stack up against the other models? It’s time to throw our contenders into the ring to find out. But before we munch on popcorn and watch the slugfest, we’ll establish some ground rules:
- We’ll use a selection of classic literature and news posts as our inputs.
- For each model we’ll generate four candidate images - in Midjourney and BestBanner this is done in one shot, in Stable Diffusion by generating four separate images.
- Since Midjourney and Stable Diffusion both have limited prompt length, we use only the first few sentences of the post as their prompt. For BestBanner we use the contents of the entire post (sometimes the plain text content, sometimes a straight-up formatted copy-paste).
- For BestBanner we’ll run one round of refinement from a selected candidate from the initial generation.
In this section we'll just showcase the images themselves. Check further down for an interpretation of how Stable Diffusion, Midjourney, and BestBanner stack up based on this benchmark, covering topicality, quality, composition, etc.
Alice's Adventures in Wonderland, Chapter One: Down the Rabbit-Hole
Stable Diffusion
For some reason, Stable Diffusion doesn't like Alice:
Midjourney
BestBanner
Astronomers observe time dilation in early universe
Stable Diffusion
Midjourney
BestBanner
Just Stop Oil funder warns more high-profile sports will be on group’s hit list
Stable Diffusion
Midjourney
BestBanner
Martin Luther King Jr.'s 'I Have a Dream' speech
Stable Diffusion
Once again, Stable Diffusion chokes. Likely due to some outdated terms in the text that may be considered offensive today.
Midjourney
BestBanner
Everything to Know About the New ‘Harry Potter’ TV Series Adaptation at HBO Max: From the Show’s Concept to J.K. Rowling’s Involvement
Stable Diffusion
Midjourney
BestBanner
茶馆 (Teahouse) by Lao She
Stable Diffusion
Here we can clearly see that Stable Diffusion can't cope with languages other than English.
Midjourney
BestBanner
Erdogan unterstützt Nato-Beitritt von Schweden - unter einer Bedingung (Erdogan supports Sweden's NATO entry - on one condition)
Stable Diffusion
Midjourney
BestBanner
Barbie Vs. Oppenheimer Is Hilarious - But 2023's Huge Box Office Battle Has A Deeper Meaning
Stable Diffusion
Midjourney
BestBanner
How does BestBanner compare?
I think it's fair to say that BestBanner wipes the floor with Stable Diffusion – admittedly, it's an old model, but unless you're going for a high-octane nightmare fuel vibe for your banners, we wouldn't recommend it.
Overall image quality: Both Midjourney and BestBanner shine here, creating detailed appealing imagery. Stable Diffusion is more about that high-octane nightmare fuel vibe.
Topicality: Stable Diffusion wobbles off-point a lot, showing vague abstract blobs when we used it to create images for the time dilation post. Midjourney is a lot better, but often misses the point (did it connect Trump to the "orange" in the text of the Just Stop Oil example?) or completely omits key information (in the Barbie-Oppenheimer example, most images are missing Oppenheimer altogether. We also see this in the MLK example. BestBanner stays on topic very well, with the Just Stop Oil example showing actual sports playing grounds (which are the scenes of protests), while Barbenheimer includes someone who actually looks like (a slightly more handsome) Oppenheimer:
Composition: Stable Diffusion quite often generates several pictures inside one image, for example in the case of Barbie/Oppenheimer. This doesn't happen in Midjourney, though (again in Barbie/Oppenheimer) we get several shots of "Person A just standing opposite Person B". The composition of BestBanner's Harry Potter images stands out, with the author of the works standing in what appears to be a posh school library.
Languages: Stable Diffusion typically falls down and creates generic images of suburbia when it comes to anything but English. Midjourney does well, but in other tests (from this German article) it interpreted "grads" as graduates, not degrees, while BestBanner worked as expected for German, Chinese and other languages.
Censorship: Stable Diffusion is very censor-happy. In some cases we can understand this, as in the MLK example where some language is used that would be considered offensive today. Quite what it has against Alice in Wonderland is beyond us. While Midjourney didn't get censor-happy in our testing, we've often seen heavy-handedness and highly-selective political censorship. In our testing with BestBanner we haven't yet found examples of censorship, though testing obscene content is way outside the scope of this post.
UI: Midjourney limits you to a Discord bot, and while there are posts saying an API is "in the works", we're not confident this will come any time soon. Several front-ends are available for Stable Diffusion, and it's hosted in many places with an open API. BestBanner also offers an easy-to-use API for generating images, and integrates with Jina AI's Python and JavaScript packages.
Overall, BestBanner stacks up well against the competition. While image quality is on par with Midjourney, in many ways (composition, topicality, language support, API) it's a superior choice. And we think the less said about Stable Diffusion the better!
Get started
Visit bestbanner.jina.ai to sign up, and join our Discord channel to share the banners you’ve created!