Image Generation

Today's Stories

1. 1

   Adobe adds AI Assistant Editor to Photoshop

   Adobe is rolling out an AI-heavy update for Photoshop, including a new "optional" interface called the "AI Assisted Editor" view. It is launching in beta and shows all AI features in a single toolbar, such as prompt-based editing, background remover, and an AI image extender. Adobe is also adding a "markup" feature that lets you draw on an image to tell the AI assistant what to change, like recoloring areas or sketching arrows. This reduces the need for text prompts, making AI editing more visual and accessible.

   The update includes an enhanced "Instruct Edit with Masks" feature. Adobe says its Firefly Image 5 model now understands the entire image context, so you can describe changes like "opening closed eyes" without manually marking areas.

2. 2

   Apple's Luce AI turns photos into relightable 3D assets

   Apple researchers introduced Luce, a 3D representation that combines geometry and PBR materials into a voxelized multimodal Gaussian cloud. It generates 3D assets from a single image, using a rectified-flow transformer to create a latent representation. Luce achieves state-of-the-art single-image-to-3D generation on the Toys4K benchmark, improving FID by 28% over the strongest baseline. It also outperforms the best baseline on a new benchmark of AI-generated images, with a CLIP image-alignment score of 0.8519 vs. 0.8299.

   Luce preserves fine details like text, logos, and inscriptions, and it can output either relightable PBR Gaussians or an optional textured mesh with tangent-space normal map. This could make AI-generated 3D assets more usable in standard rendering pipelines.

3. 3

   AI fakes are ruining trust in online animal content

   AI-generated animal videos and pictures, from polar bear rescues to found pets, are so prevalent that viewers no longer take animal posts at face value. One lost cat owner received a scam photo that ChatGPT confirmed it had generated, nearly identical to her original poster. The flood of fakes frustrates viewers and harms authentic creators like We Animals, which now faces AI accusations despite banning the technology. Experts like Oscar Horta worry that synthetic rescue videos will lead people to question real rescue efforts and hamper fundraising, especially as extreme weather events become more common.

   New laws in California and the EU require popular AI image generators to embed invisible tags, and social platforms must label AI content. Integrating verification tools into messaging apps and browsers, plus updates to model guidelines as proposed by Jeff Sebo, could help restore trust.

4. 4

   Netflix's Hamnet AI Accusation Explained by History

   Netflix's upcoming film 'Hamnet' has been accused of using AI due to an apparent error in a promotional image, but the mistake actually has a historical explanation. The accusation highlights how quickly audiences assume AI involvement in visual content, even when the apparent flaw is rooted in historical accuracy rather than technological failure.

   Whether Netflix will clarify the historical basis for the image or address the broader public concern about AI in film production.

5. 5

   Stability AI raises $76M from music and gaming giants

   Stability AI, the startup behind Stable Diffusion, raised $76 million in Series B funding, bringing its total fundraising to $232 million. The round was announced Tuesday. The investors include Universal Music Group, Sony Music Group, Warner Music Group, and Electronic Arts, as well as AMD Ventures and Pacific Alliance Ventures. This is less a typical venture round and more a lineup of companies Stability depends on for licensing and distribution deals.

   Stability plans to use the money to build out its creative production suite and expand its professional services arm. The company recently had a legal win in the UK, but a similar Getty Images lawsuit in the U.S. is still ongoing.

6. 6

   Apple Researchers Unveil STARFlow2 for Unified Multimodal AI

   Apple researchers have introduced STARFlow2, a new model built on the Pretzel architecture that combines a frozen pretrained VLM stream with a TARFlow stream. It is designed to generate and understand interleaved text and image sequences in a single, continuous, and purely causal pass. Existing multimodal models are fragmented, often sacrificing visual fidelity or degrading pretrained understanding. STARFlow2 aims to solve this by treating autoregressive normalizing flows as autoregressive Transformers, which share the same causal mask and KV-cache mechanism as LLMs, enabling a unified approach that preserves understanding while enabling high-fidelity image generation.

   The researchers report strong performance on image generation and multimodal understanding benchmarks, suggesting autoregressive flows could be a viable foundation for unified multimodal modeling. More details on the model's architecture and performance are available in the research paper.