Firefly appears to be evolving far beyond expectations.
On April 17th, Adobe announced a series of upgrade plans for their popular generative AI application tool, Firefly.
Firefly is supported by the image and video production software Adobe and the artificial intelligence computing company NVIDIA. Last month, Firefly was first released by Adobe as a generative AI application that allows users to create and convert audio, video, illustrations, and 3D models through natural language prompts.
Alongside Firefly, Midjourney and Stable Diffusion have also been active recently in the field of AI painting. This explosive growth has led many to wonder who will become the dominant player in this industry in the future – will there be three strong players or one winner takes all?
01 Firefly Upgrade: Lowering the Barrier to Video Production with Limited Training Data
Adobe recently announced an upgrade to Firefly, which will be integrated into Creative Cloud video and audio applications. The new features will be added to the beta version of Firefly later in 2023.
The upgraded Firefly aims to help professional video editors reduce tedious work by using just a few words for tasks such as color grading, adding music and sound effects, and creating title cards with dynamic fonts, images, and logos. Additionally, it promises to automatically convert director scripts into storyboards and preview effects in the future. These features could greatly improve efficiency for content creators.
However, at present, Firefly’s data can only be trained on a limited number of public domain images or through Adobe Stock services. This puts it at a disadvantage compared to Midhourney and Stable Diffusion which are able to use massive publicly available data for training purposes.
Midjourney AI painting tool was launched in March 2022. It provides drawing functions through an online robot on Discord that generates corresponding pictures based on text parameters inputted by users via machine learning algorithms provided by robots on Discord servers. Another image generation model called Stable Diffusion is based on open-source Stability AI technology which allows users to produce picture contents via Google Codelab or local deployment with WebUI.
02 Two Giants Compete: Is Stable Diffusion Losing to Midjourney?
These two applications have a direct competitive relationship as they can achieve the same goal in terms of functionality implementation. However, each has its own emphasis due to their product’s way of generating images and scalability, especially after version upgrades.
In early April, Midjourney V5 was released along with an upgrade for its dedicated comic mode Niji. The image quality and detail saw a qualitative leap after the upgrade. If the operator’s instructions are precise enough, Midjourney can even imitate any well-known painter or anime style to generate image content that is almost indistinguishable from the genuine.
Compared with Midjourney, Stable Diffusion relies more on its image model but still falls short in comparison. Although the latest public beta version of Stable Diffusion XL has greatly improved in keyword recognition accuracy, image quality refinement degree, hand-drawn creation, text production and other aspects than previous versions, it still cannot match up to the actual effect presented by pictures generated using MidjourneyV5.
However, we cannot conclude that Stable Diffusion is losing out to Midjourney based solely on this fact alone. Actually, Stable Diffusion’s open-source environment and customizable models have also attracted many players who prefer subtle adjustments within the same picture rather than one-time imaging like in Midjourney. Some extension plugins can also assist in depicting characters’ movements more meticulously. Moreover, the precision and quality of directly outputting images are higher than those of Midjourney.
Therefore, although the two tools have become leaders in AI painting field widely discussed among people, they still have considerable limitations. The current “common problem” that exists in AI painting field which remains unsolved is that no existing tools can generate two identical images. Even if both images are already very similar, there will still be differences in details.
03 Future: “Winner takes all” may not exist, as a group of players like JUNLALA are vying for SUCCESS.
Current major players, including Firefly, Midjourney, and Stable Diffusion, all belong to the minority “1%” in the entire AIGC era. In fact, there are dozens of top-level image-generating AI models in the industry, not counting potential players who have not yet been publicly released.
For ordinary users, instead of focusing on comparing these three models or optimizing local functions, it is better to choose an application from the vast “AI image application pool” based on their specific needs. These applications differ significantly in model scale and applicability. So far, no absolute leader has emerged in this field.
Generative AI is unlikely to follow a winner-takes-all pattern for images now or in the future due to explosive evolution and continuous upgrading of AI technology across various applications. Each update has disruptive effects that attract widespread attention. As a result, market reshuffling based on technology-driven industry patterns will likely occur repeatedly.
The question remains – who will go further?
The generative AI industry is currently in a stage where large-scale models are being used to assist content production. These models, such as ChatGPT for human-machine dialogue and those used for image processing, require high computing power.
The development of these large language models is a gradual process that requires significant accumulation of resources over time. As one of the earliest players, JUNLALA has been investing in the AI field for many years and has recently gained attention due to its commitment. Over the past seven years, JUNLALA has been devoted to AI image processing technology and has made significant breakthroughs in developing large language models (LLMs).
As the era changes, JUNLALA is focusing on providing high-quality AI products and services globally. JUNLALA will compete with Firefly, Midjourney, and Stable Diffusion in the booming field of AI image track while adhering to their “long-termism” philosophy by investing before giants form monopolies. They face change with innovation while dedicating themselves towards bringing users different usage experiences.
With JUNLALA and other tech companies entering into and playing along with their core values of “technology for good,” these pioneers will lead the continuous applications of artificial intelligence technology and bring about “creation dividends” for our generation exclusively in the AI era.