Image Generation

Today's Stories

1. 1

   Adobe (ADBE) and Broadcom (AVGO) highlighted as AI stock picks

   An undervalued-stock screen highlighted Adobe (ADBE) and Broadcom (AVGO) as AI-driven picks, flagging Adobe's generative AI credit usage as a driver of Average Revenue Per User (ARPU) expansion. As professionals use more generative AI in Adobe's tools, they may hit plan limits and upgrade or buy credit packs, lifting spending per user.

   A single shift in assumptions about how aggressively enterprises pay for deeper Adobe AI usage could reshape its risk and upside, so read the full narrative.

2. 2

   G-SHOCK taps AI plus designers to turn concept into product

   Casio's G-SHOCK team used AI image generation alongside its own designers to develop a watch concept into a producible product, with the designers recounting the commercialization process. It shows AI acting as a co-creation tool in physical product design rather than replacing the designer, with people still shaping the final output, according to the designers.

   The test is how much of the concept survived the move to a manufacturable product, and whether this way of working spreads inside G-SHOCK's lineup.

3. 3

   Black Forest Labs' FLUX 3 Action sets RoboLab-120 record

   Black Forest Labs released FLUX 3 Action, an open robotics model that takes multi-camera video feeds and predicts a robot's next action. BFL says it sets a RoboLab-120 success-rate record with seven billion parameters, up to 3.95 times faster than the previous best open model. The result suggests strong robot performance may not require huge models, which BFL says matters for running on a robot itself rather than off-board. That could widen who can build and test robotics AI, since the weights are free to download.

   The record is BFL's own claim tied to one benchmark, so the test is whether FLUX 3 Action holds up on other tasks and real hardware. BFL also points to video games as a near-term testing ground, with computer-operating agents as a possible later step.

4. 4

   Apple's probe guidance sets new state of the art

   Apple researchers introduced probe guidance, which uses frozen internal states of an existing diffusion model to guide flow matching models; on continuous diffusion language models it set a new state-of-the-art on unconditional generation. The method eliminates an extra forward pass at inference time and reliably ensures weak and strong models share similar dynamics, which could make diffusion language models more practical.

   The finding that the weak model must come from a low-entropy region of training may shape how future autoguidance setups are built; watch adoption of probe guidance in diffusion language model research.

5. 5

   Oracle adds graph-aware retrieval and image memory to its AI Agent Memory

   Oracle announced new features for Oracle AI Agent Memory: graph-aware retrieval, image memory, and enterprise controls. These additions suggest Oracle is widening what kinds of data agents can use, including images.

   Enterprise controls point to governance needs, but the announcement does not detail availability or pricing.

6. 6

   Invideo: GPT‑6 Astra triples color-grading success

   Invideo's CEO Sanket Shah says GPT‑6 Astra lifted the success rate of color-grading and color-correction tasks about three times, and a few editors built about 50 custom effects in one day. Color work is where invideo previously saw very high failure rates, so this gain points to AI agents handling more of the tedious frame-by-frame planning in professional editing.

   The gain is a success-rate comparison versus earlier invideo tests, not a finished product claim. Watch whether Astra's fewer reasoning steps and output tokens hold up on longer, more complex edits.