You’ve been chargeable for Managing the ARTA – AI Art generator from the ideation phase until now. Could you share some insights on these early days?
After all! Those were dynamic times. We managed to release a finely made application inside just every week, becoming one in all the primary consumer app creators to supply text-to-image generation functionality on mobile. Our goal was to construct a mass-market product providing individuals with “an artist” of their pocket. So, because the conceptualization and early development stages, we’ve taken a concentrate on usability and scalability. But despite entering the market very timely, it was quite difficult to grow our install volumes to an adequate extent, even with a superb media buying team like ours. A big boost occurred three months after the app’s release when our Avatar feature got hyped. The amount quickly became moderately high for our area of interest, and since then, our task has been to keep up and increase it.
What was the unique tech stack that you simply launched on and what were among the challenges with art generation during this era?
We launched based on Stable Diffusion 1.3 using the official API from Stability.ai. I should say the situation with the standard of generations then and now could be like night and day. Once we first began, our QA managers often reported issues related to the aesthetic value of images or inaccuracies in representing specific concepts and features. Nonetheless, that was standard for Stable Diffusion at the moment. Now, generation output is significantly better in all elements, including stylistic reproduction, composition coherence, visual fidelity, level of detail, and more.
Shortly after the app’s release, we began renting servers on Amazon, and supporting them turned out to be quite a challenge. Even with sufficient funds, there could also be no free A100 available once you need it, and you should have to attend for a few days. Thereby, we needed to live without autoscale, redirecting all excess traffic to our partners’ APIs.
Maintaining all of this stays moderately tricky to at the present time, with minor issues occurring on one end or the opposite every month or so. For instance, we occasionally encounter temporary problems with the standard of generations when the provider updates the server, tests weights, or implements other changes that affect the generation output. Such errors can last from an hour to half a day and are unpredictable and difficult to trace. Normally, by the point our support department receives a user report about blurry images or another occurring issue, the API provider has already fixed the issue. Nonetheless, it’s a serious concern for our users. Due to this fact, we at the moment are constructing a system that mixes multiple providers and our own servers for special generations, allowing us to have more control on our side of things.
As a product manager, what strategic decisions have been pivotal in guiding ARTA to its top-ranking position shortly after its release?
ARTA’s (at the moment called Aiby) early rise resulted from the timely decision to implement the viral Avatar feature when it just began making rounds on social media. We quickly recognized the growing interest on this functionality. Our entire team, including product, marketing, and development, was on the identical wavelength and visionary about its success. We also acknowledged that a short while to market was crucial. So, from day one, we dedicated all our resources to realizing this feature, prioritizing it above other tasks.
Since our deadline was ASAP in order to not miss the moment when AI Avatars reach their hype peak, we opted to make use of a third-party solution and customize it for our app. While avatars were starting to realize traction on mobile, the technology had already been available on the internet for a while, even with an API. Due to the team’s concentrated efforts, our first working version was within the App Store in only five days, offering highly competitive avatar output. It helped us attain the #2 position within the American top charts and stay the second most downloaded app within the US for every week.
Your team has recently released an upgrade to ARTA’s AI avatar generation feature. Could you share some details regarding this?
The AI models are likely to add generic facial expression during training, making avatars look different from the source photos, and the more unique one’s traits are, the more unlike the AI interpretation can appear. To deal with this issue, we decided to create our own avatar service. We had been using a third-party API for a very long time but didn’t yield significant improvements. With the server shift, we were in a position to arrange more optimal training technology to higher maintain the likeness of the user’s real face within the avatar output. While I can’t disclose our unique pipeline intimately, it became possible as a consequence of a selected combination of SDXL settings, LORAs, and face enhancers, and we haven’t yet seen higher outcomes elsewhere.
With the brand new server, we moved away from a hard and fast cost for every avatar pack to a monthly server fee and might now offer avatars through a weekly subscription as a substitute of requiring separate in-app purchases. It creates a more fulfilling experience and is less expensive for our users in the event that they need to generate, for instance, five avatar packs inside every week or change the photo input as they go. Considering all the above, our avatar offer currently boasts the very best price-performance ratio available on the market. While there are apps capable of making high-quality realistic avatars, ARTA stands out by providing a various range of vivid and colourful output variations besides realistic styles, all with the identical precise level of facial recognition.
In what other ways has the team improved the app’s capabilities?
We concluded that using third-party APIs is more efficient for common use cases like text-to-image generation, image conversion, and inpainting. This approach eliminates the necessity to spend time determining the best way to integrate these functionalities into our server infrastructure. Moreover, it reduces costs in situations when a brand new feature doesn’t take off as expected and we determine to remove it. The AI image generation industry is rapidly evolving, with quite a few dedicated services available, so we explore and regularly adopt people who align with our objectives.
At the identical time, ARTA’s needs often become quite unique, requiring in-house findings. In cases when tailored APIs are either non-existent or don’t provide satisfactory output quality, we specialize and customize our internal services and develop our own solutions to attain the outcomes we wish. For instance, along with upgrading AI Avatars, our ML and prompt engineers have give you a brand new pipeline for the app’s AI Filters (Selfies) feature. We’ve also developed a singular algorithm for our upcoming AI Baby feature – a generating functionality that enables two people to merge their photos and see how their child might look. Based on my perception of the world as a product manager, I initially doubted its success, but ad creatives featuring this idea are very talked-about. So, checking up on marketing insights is very helpful in content-related cases.
Can users influence the artistic process in ARTA? In that case, what tools and options can be found for users to customize the AI-generated artwork?
We handle all of the complex elements related to generation, aiming to offer our users with an easy artistic experience without unnecessary technical overload. So, the first way users influence the output is thru prompts. We keep this process transparent by showing the precise word request that might be sent to the model for generation and only offer assistance with composing effective prompts if needed.
We select the very best default settings for every integrated model so users don’t hassle about that. Typically, there’s no need to regulate them to maximise results, as they already produce an optimal generation output. Still, if the user desires to experiment, a complicated mode is one tap away, and a few deeper parameters are within the settings section.
Soon, we are going to add a Seed parameter, allowing users to have complete control over generation after they have to recreate the same image from scratch. Moreover, we plan to expand the list of aspect ratios. We’re also pondering of adding several controlnets to regular generations. They’re already supported on the server side, as we use them to generate AI Filters and sketches, but they aren’t yet delivered to finish users.
How do you perceive the impact of AI like ARTA on the normal art market? Do you see AI art generation as a disruption or an enhancement to the art industry?
I see it as an enhancement. Generative AI has introduced latest and invaluable opportunities to boost the artistic process while significantly reducing turnaround time. It assists digital artists, designers, illustrators, and other visual content creators with a wide range of tasks, from exploring ideas and developing concepts to generating sketchups and ready-to-go images. Ultimately, our ability to leverage its advancements is just limited by our imagination.
For instance, I actually have a hobby of making PC games, and recently, I used ARTA to generate a set of icons for skills and items. I could design them alone using Adobe Illustrator, but with a picture generator, I got what I needed almost immediately. My wife, in turn, is a retoucher-photographer. Due to Photoshop’s Generative Fill, she works much faster and has more free time (or more income if she decides to just accept more retouching orders).
When done well, AI-generated images can look indistinguishable from skilled artwork. Nonetheless, in my view, AI won’t ever replace a real skilled. Regardless of how expert neural networks change into, they’re still trained on data created by humans, meaning that every thing they generate already exists somewhere. As then and now, truly revolutionary ideas can only be produced by people. While the normal meaning of art stays related to human-made pieces, AI art is like an anticipated spinoff, inviting everyone, no matter artistic background, to try an exciting latest experience.
Looking beyond just improving image quality, where do you see the long run of AI image generation heading?
Together with the image quality, the speed of generations will increase, mechanically resulting in less expensive outputs.
I feel it won’t be long before there’s a straightforward solution to generate the identical characters in several environments and positions so that we’ll see the rise of AI in comics, kid’s books, game graphics, and more. Interior design and ad creatives production are already the spheres actively leveraging generative AI, but more is ahead of us because the technology continues to evolve.
Considering that every one generations require strong GPUs, these technologies will develop together with AI for quite a while. We’re only yet initially of the journey. Perhaps the brand new Apple of our time might be Nvidia, with everyone, or no less than those within the IT industry, anticipating latest video card releases just as all of us did with iPhones.
AI image generators will proceed delivering fun and fascinating experiences, whether by introducing latest concepts emerging from popular culture or reviving older ideas enhanced with higher technology. For instance, interest in AI Baby generations is currently growing. One recent technology based on Stable Diffusion has demonstrated impressive output from merging two individuals’ features to disclose their biological child’s potential appearance. The outcomes far surpass what was available on horoscope sites a number of years ago, and persons are wanting to give it one other try.
What are your predictions for what we must always expect next from Generative AI?
The wave of recognition for video generation is on the horizon. With advancements in technology reaching a sufficient level, there’ll undoubtedly be attempts to coach neural networks using people’s facial expressions and gestures to create video avatars, potentially even with unique user voices.
AI Audio is one other significant breakthrough ushering in a brand new era for the music production industry. This technology has already presented amazing opportunities for composing songs based solely on text input, making it a superb tool for creating custom non-stock soundtracks for various kinds of video content. Overall, it’s really fun to take heed to something as mundane as Terms of Use rapped or sung with romantic intonation.
Thanks for the nice interview, readers who want to learn more or generate some images should visit ARTA.
