Draw It and It Appears: Scribble-to-Object Editing Comes of Age
A small research LoRA lets you scribble over a region and rename the object. It is a clean sign that the work in image AI moved from generation to editing.
Blog
Discover tips, tutorials, and insights about Generative AI
Re-shooting Reality: Seedance 2.5 and the Camera That Doesn't Exist
Seedance 2.5 can re-shoot existing footage from a new camera angle. That changes coverage, continuity, and what footage can prove.
When the Assistant Runs the Office: Gemini, Agents, and the New Workflow Layer
Google's new work agent coordinates tasks across Gmail, Drive and Calendar, and can route some work to Claude. The orchestration layer is the product.
The Sub-Agent Economy: Why Cheap Models Are the Real Agent Story
Claude Haiku 5.5 cut running costs by about 75 percent and was positioned as a sub-agent. The cheap tier is where agent economics actually change.
ChatGPT Stopped Answering and Started Building Tools
GPT-6's Intelligent UI lets ChatGPT compose charts, forms and working tools instead of writing paragraphs. That changes who designs the interface.
No Green Screen: Alpha Gen and the Quiet Revolution in Video Keying
LTX's Alpha Gen pulls a clean matte from ordinary footage with no green screen. The hard edges it targets are where the whole problem lived.
Brand Kits Are the Boring Feature That Makes AI Design Usable
Magnific One and ElevenLabs both shipped brand kits within a day of each other. The boring feature is what makes generated images usable.
Vibe Video: The Coding Model That Makes Films by Writing Frontend Code
The most talked-about AI videos this month were made by a coding model writing frontend code, not by a video generator.
The Two-Frame Method: How First and Last Frame Interpolation Changed AI Video Planning
Veo 3.1's first and last frame feature puts two decisions back at the front of AI video: where a shot starts and where it ends.
After One Yuan Per Second: The Cost Ledger for AI Short-Drama Creators
When generating a single short-drama episode costs 400 to 500 yuan and saves are being discounted, what AI content creators face is really a very concrete cost ledger.
The Memory Chip Is Now the Bottleneck in AI Compute
The logic roadmap is public and predictable. The memory roadmap is gated by packaging yields, a years-long talent base and three companies' capital plans.
Satellite Photos Are Now Robot Training Data
The bottleneck in physical AI training stopped being compute or model capability. It became the quality of the synthetic world.