New AI models from Adobe, called Firefly, can produce editable vector drawings, crisp photos, etc
The current version of the Adobe AI model was trained on roughly 70% more photos, according to Alexandru Costin, vice president of generative AI at Adobe. This should increase the system’s capacity to produce photo-realistic images and bring it closer to the output of rivals.
New controls and prompts for the AI generation, including uploading a picture to mimic its aesthetic, will also be included in the most recent edition. To safeguard artists from infringement, Costin stated that Adobe will maintain an internal log of the reference photos that are uploaded. Branding Firefly as “commercially safe,” the firm promises to support clients who are the target of copyright lawsuits.
The long-standing leader in creative software, Adobe, is attempting to outpace companies like OpenAI by incorporating the new technology into industry-standard products like Photoshop. It is doing this by constructing AI models that are mostly trained on its extensive stock picture library. Adobe raised the cost of its product last month after releasing the new AI tools to the general public.
With the new AI model, vectors—scalable computer visuals that are common in digital design—can now be generated in Adobe Illustrator program using a text prompt, the company announced in a statement on Tuesday. According to Costin, the stock library of Adobe contains “tens of millions” of vector images, which are used to train this model. An upcoming application would enable users to design posters and other marketing materials, which are typically constructed using a template in Adobe Express, a program that rivals Canva Inc., an Australian firm.
According to the company, those who submit content to Adobe Stock and have it utilized to train artificial intelligence models receive an annual bonus. Stock contributors won’t be compensated extra if their photos are used in the training set of many models, according to Costin, because the payment mechanism is dependent on the intake of fresh data.
