Adobe has recently retrained its generative ai, and i must say, i'm genuinely impressed with the upgrade. i decided to put adobe's firefly image creator to the test, and while it's not flawless, it's undoubtedly a wellspring of creative output. additionally, adobe has introduced generative ai to illustrator designs and adobe express, which is a significant development for those of us looking to stimulate our creative juices with artificial intelligence.


the previous version of firefly allowed users to transform text prompts into images on the firefly website and adobe's photoshop image-editing software. however, with this second-generation ai model, things have taken a leap forward. the new model, according to alexandru costin, adobe's generative ai leader, offers more intricate details, improved image quality, and higher resolution. users can even take advantage of photography parameters like lens focal length and depth of field to influence the image generation process. although the new ai model is not yet available in photoshop, it is expected to undergo testing online before its official release.


during my testing, i noticed a remarkable improvement in both the results and the user interface. adobe trained the new ai model using double the number of images, resulting in more lifelike and detailed output. as a user, you'll appreciate the enhanced quality, including the ability to see minute details like skin pores. moreover, the improved model empowers you to steer the generation process with greater precision by adjusting various parameters.


generative ai is built upon the same core principles as traditional ai methods, leveraging machine learning to recognize patterns in real-world data. however, generative ai takes things a step further by enabling the creation of entirely new materials such as text, images, speech, or video based on its understanding of those patterns. this groundbreaking development has unleashed the true potential of computers, elevating them from their literal, predictable nature and providing us with a glimpse of what a genuinely intelligent machine can achieve.


now, let's delve into my experience with adobe's upgraded firefly generative ai. despite occasional instances where the results were less than convincing, i found it immensely enjoyable to create imaginative and entertaining images, particularly when utilizing forgiving art styles like paintings, cartoons, and watercolors. the process involved entering prompts and observing what the ai produced, followed by tweaking the prompts to achieve more useful or extraordinary outcomes.


furthermore, the interface has undergone significant improvements. the inclusion of a high-level option to choose between photorealism and a more artistic illustration style provides valuable flexibility. additionally, users can select from various aspect ratios, such as square, landscape, and portrait. sliders for visual intensity and style strength allow you to fine-tune the final image, offering a range of options from dramatic to subtle. moreover, you have the option to upload a reference photo to guide the style of the output.


as one of my standard tests for ai, i input the prompt "parachuting elephant," and i must admit, the second-generation firefly produced better results than its predecessor. however, the ai still struggles with accurately representing the parachute cords. nonetheless, in one instance, it managed to construct a wooden frame to support the elephant, which was a pleasant surprise.


next, i tried generating "a kindly doctor in a hospital room for a pharmaceutical ad," and the results were generally satisfactory, with firefly offering portraits depicting a variety of races and genders. however, each portrait featured the obligatory scrubs and stethoscope.


moving on, i explored the generation of spikey electric guitar images. while the results were satisfactorily menacing, they sometimes lacked the correct number of frets, pickups, or strings.


it's important to note that, despite these positive aspects, there is still significant room for improvement. for example, in images of the stereotypical hacker hunched over a keyboard, the ai occasionally depicted them wearing their hoodies backward. moreover, my attempts at generating a photorealistic image of a red crab waving its claws in the air were unsuccessful, with distorted claws, extra eyes, and other issues. firefly did manage to generate an acceptable groundhog image, but it's clear that adobe's training data lacks sufficient information about picas, a rather unusual rodent found in high mountain regions.


another test i conducted involved generating an image of an angry crocodile leaping out of a stormy ocean with lightning striking all around. unfortunately, the ai struggled to depict a plausible arrangement of teeth. similarly, my prompt for a "ghoul in a heavy metal outfit riding a mountain bike through a post-apocalyptic wasteland" illustrated the difficulty of aligning the geometry of a humanoid character with the mechanics of a bike. lastly, my prompt for an image depicting "a high-tech communications network spanning the globe with data surging through the wires" resulted in a rather convoluted mess.


however, it is worth mentioning that you can refine your prompts to achieve better results. additionally, the art style often yields more convincing subjects compared to the photo style. some art styles, such as doodle drawing, are more forgiving in this regard. personally, i found the watercolor option particularly appealing. nevertheless, it's essential to acknowledge that firefly, despite its impressive capabilities, is not flawless.


Firefly has arrived in adobe illustrator, the favored tool for designers who create vector graphics such as logos and diagrams. similarly to its integration in photoshop, the software now offers firefly text prompt capabilities. it utilizes these prompts to generate a quartet of illustration options, which designers can choose from and further refine through editing. the resulting illustrations are fully editable, enabling greater flexibility and customization.


furthermore, the adobe express application, designed for creating materials like flyers and social media videos, now features its own text prompt field. this field allows users to build templates by inputting specific prompts. for instance, typing "pirate themed birthday party announcement for children" will generate a range of layout options complete with custom art and fonts. currently, this feature is in beta testing.


while the creativity of generative ai can pose challenges in fields requiring factual accuracy, such as law, travel planning, or academic writing, it holds immense potential for creative applications. adobe aims to leverage firefly to revolutionize its tools and empower individuals lacking expertise, allowing them to explore their creative potential.


during the max conference, adobe unveiled several ai-powered projects in its traditional "sneaks," offering a glimpse into forthcoming technology. one noteworthy project, project stardust, introduces a photo editing system capable of analyzing scenes and separating elements within them. this enables users to select, move, delete, resize, or modify specific subjects effortlessly. the system even recognizes and adjusts shadows accordingly, streamlining the process of making targeted edits within photos.


among other ai-powered sneaks showcased by adobe are project fast fill, which integrates generative ai into video editing. it allows editors to remove individuals from the background of a single video frame and extend the deletion throughout the entire clip. moreover, it can track designated areas, dynamically adjusting changes as subjects move. for instance, adobe demonstrated adding a necktie to a person's outfit in motion and applying a new foam pattern to the surface of a cup of latte.


another intriguing project, project poseable, enables the conversion of 2d photos of people into 3d models. these models can be manipulated to assume preferred poses, with users repositioning limbs and joints accordingly. additionally, with the aid of text prompts, users can create new character styles, such as ogres or nurses. while project poseable currently lacks consistency in maintaining characters across different creations, adobe envisions its potential for storyboarding purposes.


lastly, project draw & delight allows users to transform simple sketches accompanied by text labels into cartoon-like designs. users can enhance their designs by adding supplementary material through additional sketches and prompts. the system ensures that the character chosen earlier remains intact during the modification process.


firefly has proven immensely popular, with adobe customers generating over 3 billion creations in just a few months of testing. this remarkable response surpasses the typical utilization rate of photoshop features. consequently, firefly and similar technologies represent promising avenues for adobe's growth. however, there may be associated costs as adobe intends to raise creative cloud subscription prices in november by roughly 9% to 10%.


to accommodate the computational demands of these generative ai models, adobe has leveraged its cloud computing infrastructure. this infrastructure incurs real expenses, especially considering the high cost of nvidia processors, which handle most generative ai tasks. currently, these large ai models often require more memory than an ordinary laptop can provide. nevertheless, industry progress and the efforts of companies like amd, intel, and apple to enhance processor capabilities offer potential solutions to this limitation.


additionally, adobe employs ai to assist users in harnessing its creative tools. in photoshop, the software suggests completions for text prompts to create more refined results. this feature proves particularly helpful for individuals less experienced in the field of prompt engineering, ensuring both grammatical correctness and aesthetically pleasing image generation.


in lightroom, adobe introduces a new ai-based tool for generating lens blur effects, artificially replicating the bokeh achieved by high-end lenses. this effect isolates subjects by blurring the background, contributing to visually appealing and professional-quality photographs.


businesses utilizing firefly can further customize its functionality by uploading their assets. this allows them to guide the firefly generation process precisely according to their needs and preferences.