From: redpointai
AI is transforming creative tooling across images, video, music, and more [00:00:00]. Scott Belsky, founder of Behance and Chief Product and Strategy Officer at Adobe, discusses the significant impact of generative AI on creative workflows, business models, and the very nature of content creation [00:00:10].
Adobe’s Strategy and Innovations
Adobe’s approach to AI is multifaceted, focusing on interfaces, models, and underlying data [00:03:00].
Firefly Models and Data
Adobe’s Firefly family consists of homegrown generative models trained on licensed content [00:03:10]. They have a compensation program for content contributors and aim for blanket indemnification for customers [00:03:20]. A key focus is enabling custom models for customers, such as Nickelodeon training a version of Firefly on “SpongeBob SquarePants” to generate new character development and storylines without concern for compliance [00:04:06]. This ensures commercial safety and compliance, addressing a top concern for customers [00:04:41]. For Large Language Models (LLMs), Adobe utilizes partnerships, integrating them into products like Acrobat and digital experience platforms for marketing analytics [00:04:26].
User Workflows and Controls
Adobe integrates these models into existing products like Photoshop and increasingly makes them available to third parties and large companies for scaled workflows [00:03:34]. A significant insight from Adobe’s Project Neo (a 3D illustration program) was an unexpected workflow where users combined Project Neo with Adobe Firefly for “extreme precision” in image generation, demonstrating the power of user-discovered controls [00:01:15]. This highlighted that the focus should shift from the “next best model” to the controls applied on top of the model [00:02:18].
Customer-Centric Approach
Adobe builds its own models in areas where it believes it can deliver a superior end-to-end experience, leveraging its expertise in media generation, fine-tune controls, and user data [00:11:35]. For commoditized or highly specialized areas where other companies might invest more, Adobe prefers to partner [00:12:41].
Evolving Creative Workflows
Unlocking Exploration and Efficiency
Generative AI allows creative professionals to explore a far greater surface area of possibility much more quickly, overcoming the constraint of time [00:06:27]. An example is “Generative Recolor” in Illustrator, which instantly applies hundreds or thousands of color palettes to vector creations based on prompts, a task that previously took days [00:06:42]. This frees up creators for higher-order exploration [00:07:04].
The “Blank Canvas” Problem and User Education
A challenge for startups and larger companies alike is the “blank canvas problem,” where users don’t know how to start with powerful AI features [00:07:23]. Adobe’s job is to ensure users can “flex their taste” through prompt augmentation and onboarding experiences [00:09:10].
The Importance of Taste
Just as photography evolved beyond simply clicking a button to include choices of lens, lighting, and selection, taste becomes more important than rudimentary skills offloaded to compute [00:08:26]. Taste is crucial, as seen in the success of Midjourney due to its strong aesthetic choices [00:09:30].
The Landscape of Generative Models
Convergence vs. Diversification
Initially, there was a belief that only a few major models would dominate, but the reality is likely thousands of models, many becoming commodities or focusing on niche use cases with specific tuning and data [00:09:56]. Tools should allow users to choose different models for different purposes [00:10:17]. For example, using models trained on diverse content for ideation and mood boards is acceptable, but for commercial purposes, commercially safe models trained on licensed data are essential [00:10:29].
Commercial Viability in Generative Video
The generative video AI space is experiencing rapid growth with many startups, but the bar for “professional grade” and “commercially viable” media is extremely high [00:14:17]. There’s a significant gap between “cool happy path demos” and pragmatic, everyday quality use cases [00:15:02]. A key milestone for video models could be enabling minor extensions of existing scenes (e.g., adding a few seconds) to save significant re-shooting costs [00:15:28].
Balancing In-house vs. Partnership Models
Adobe’s models are built when they can deliver a better end-to-end experience by leveraging their expertise and data [00:11:35]. However, for commoditized areas, partnering with other companies is the preferred strategy [00:12:41]. Adobe views external models like Pika and Sora as platforms that enhance their own products [00:30:19].
Hyper-Personalization and its Implications
Flooding the Zone with Content
The creative and marketing space is moving towards hyper-personalized digital experiences at scale [00:17:22]. Websites and emails will be tailored, featuring videos based on past purchases, leading to an inundation of personalized content [00:17:34]. Future generations will expect this level of personalization [00:20:29].
The Craving for Scarcity and Craft
As AI makes content free to generate and easily varied, translated, and localized, humans will increasingly crave scarcity, meaning, and craft in their digital experiences [00:20:50]. This may lead to the concept of “luxury software” and a reboot of the role of humans in content creation [00:21:14]. Shared social experiences, like discussing entertainment, are unlikely to disappear [00:22:07].
The Future of Music
In music, while AI can lower the floor for participation, the human element of storytelling behind a song is crucial for resonance and repeated listening [00:22:36]. AI tools must not miss this human connection between artist and consumer [00:23:28].
Future Capabilities and Business Models
Proactive AI Suggestions
Beyond simply performing tasks or answering questions, the next tier of generative AI will involve proactive suggestions [00:24:25]. This includes agents suggesting new creative directions or predicting performance in different geographies [00:24:30].
Speed, Quality, and Inference Costs
Customers universally desire increased speed and quality [00:24:56]. While inference costs are a consideration for efficiency, they do not constrain Adobe’s innovation in unlocking valuable customer experiences [00:25:23].
Generative Credits and Pricing
Adobe uses a generative credit system [00:26:01]. Customers receive a basic amount with existing plans, with options to upgrade for more [00:26:48]. This model allows flexibility for more intensive future capabilities like generative video, where credit costs can be adjusted while maintaining accessibility as a core mission [00:27:10].
Insights and Outlook
Biggest Surprise: The Power of Defaults
The biggest surprise in building AI features was the impact of placing “Generative Fill” and a generation bar as defaults in Photoshop [00:28:11]. This “unlocked utilization in a way we could have never forecasted,” highlighting that AI needs to be a default part of workflow to give users superpowers [00:28:33].
Changing Mind: The Future of Models
The initial thought that only a few models would dominate has changed; instead, there will likely be thousands of models, with much of the activity happening “on the edge” [00:29:07]. As model capabilities increase rapidly, many use cases fall below the “frontier,” making cost a more significant focus than choosing the single “best” model, especially with the rise of local or open-source models [00:29:31].
Startup Opportunities
Significant startup opportunities exist in areas where AI can make cumbersome processes efficient [00:18:49]. This includes:
- Small Businesses Operating as Huge Businesses: Enabling small businesses to perform marketing and management at a scale previously reserved for large enterprises [00:19:06].
- Revolutionizing Antiquated Industries: Tackling spaces like law or government where processes are inefficient [00:19:29].
- Personalization in Physical Spaces: Extending hyper-personalization from digital experiences to physical interactions, such as feeling “known” at a restaurant or store [00:32:36]. This involves unlocking the “personal driver” sensation of Uber for hospitality, where AI can scale experiences previously only available to a select few [00:33:09].
Scott Belsky's Newsletter
Scott Belsky shares his thoughts on these topics in his monthly newsletter, “Implications,” as a way to connect dots and provoke thought among his teams, founders, and peers [00:34:15]. He emphasizes that he does not use AI in its writing [00:35:00].