New Framer's biggest update yet puts AI agents directly into the design canvas and adds team branching and a community marketplace, which together change how a site goes from idea to live.
New A fine-tuned LoRA on FLUX.2-klein that takes a black-and-white manga page and a reference image and colors the line art to match the palette and style of that reference.
New A research paper that teaches a model to read a single human-centric photo and spit out three cinematically reasoned crops, the kind of establishing-medium-close-up sequence an art director would normally plan by hand.
New A conditioning framework that lets you specify fonts in generated images without the usual tradeoff where tighter typographic control makes the actual letters fall apart.
New A research method that takes a static font and predicts the glyph variations needed to make it fully variable, covering weight, width, slant, and more in a single model.
New Roblox released open weights for a diffusion model that turns text into 3D meshes, which puts game-ready asset generation in reach for anyone with the hardware to run it.
New Describe a campaign in plain English and Brew builds the copy, layout, audience targeting, and send logic in one pass, which saves the usual back-and-forth between a writer, a designer, and your ESP.
New Google's Stitch generates mobile and web UI screens from text prompts on a live canvas, then hands them off to Figma or your builder of choice in one click.
TASTE is a preference dataset built by professional designers, and it shows that the AI judges most teams rely on today do not actually agree with how designers see the work.
A research method that converts raster images into SVGs by rethinking where anchor points go, so the result is actually editable, not just technically vector.
PIXLRelight takes a photograph and a lighting specification and hands back a relit version in under a tenth of a second, using physically based rendering intrinsics to make the light actually behave like light.
A research framework that treats individual hair strands as the generative unit, giving you compositional control over 3D hairstyles by scalp region and strand hierarchy.
VAnim is a research framework that takes a text instruction and moves SVG elements through it, without the structural damage that usually comes with that territory.
A free, local desktop app that turns text prompts into web, mobile, and slide prototypes using whichever AI coding CLI you already have set up.
A rules file you drop into Claude Code, Cursor, or Codex that steers the model away from the visual clichés that make AI-built interfaces look generic.
A design agent that builds working prototypes using your existing styles and components, then hands the result off to engineers without the usual back-and-forth.
Anthropic's design-focused Claude mode lets you talk your way to a working prototype, slide deck, or one-pager, which is useful when you need something real on screen fast.
A tool that lets AI coding agents read your actual Figma design system, so what they build matches your tokens and components instead of inventing their own.
A set of 19 skills that wire Claude Code into Higgsfield AI, letting you drive image generation, Seedance 2.0 video, and full UGC ad pipelines from a single conversational interface.
Feed it your visual identity and Hue produces a full design system, so any UI your AI assistant builds comes out on-brand without manual cleanup.
A Claude Code skill that bakes Material Design 3 rules directly into AI-assisted coding, so generated components stay on-spec without you chasing down the guidelines yourself.
Nvidia's Lyra 2.0 takes one photo and hands back a 3D asset, which cuts out a lot of the manual work that usually sits between a reference image and something you can actually use in a scene.
A set of skills for Claude and similar agents that wire in a structured design process, so the AI follows a deliberate path instead of throwing output at the wall.
Lokuma sits between your coding agents and the final output, giving tools like Claude Code or Codex something closer to actual design thinking when they build pages.
DancingBox turns the movement of any physical object you wave in front of a webcam into a smoothed, realistic 3D character animation, with no motion capture rig required.
A Claude Code skill that routes image generation through Gemini, so you can prompt, generate, and iterate on visuals without switching out of your coding environment.
Cardboard works through your footage as an AI collaborator, making editing decisions and assembling a cut so you spend your time refining instead of roughing out.
Chronicle turns your notes or a rough prompt into a branded presentation draft, then lets you keep refining it through conversation rather than hunting through menus.
67 ready-made DESIGN.md and SKILL.md files that tell agentic tools like Claude Design, Cursor, and Codex how to handle design work the way you actually want it done.
A Figma plugin that pulls your local styles and variables out of a file and writes them into DESIGN.md and SKILL.md, so Claude Code, Cursor, or Codex can reference your real design system when it generates UI.
A research framework that paints UV-space garment textures guided by a character reference image, keeping the result consistent across the full 3D surface without needing the reference to line up with the mesh.
A set of Claude Code skills that gives the AI enough context about ComfyUI's node APIs to actually write useful custom node code, rather than hallucinating its way through the framework.
Kodo takes a text prompt and returns a layered, editable design — poster, slide, social graphic — so you spend your time adjusting, not rebuilding from scratch.
TypeUI gives AI coding agents like Claude and Codex a shared design vocabulary, so the UI they generate actually looks consistent instead of made-up on the spot.
Anthropic's Claude now lives inside PowerPoint itself, reading your templates and making edits in real time so you stay in one place instead of bouncing between a chat window and your deck.
A plugin that connects Claude Code to your ComfyUI instance, so you can generate images and video, edit the graph, and manage models all from a conversation.
Inspector sits between your browser and your codebase, so you can click an element, tweak it visually, and have the change land in the repo without touching your AI agent's chat window.
A ComfyUI custom node that exposes Anima's anime and illustration generation as an API, so your AI coding agents can call it directly instead of you doing it by hand.
A research framework that edits graphic design layouts geometrically while holding the untouched elements in place, and does it without needing matched before-and-after training pairs.
A research system that builds full icon grids by letting designers navigate both what an icon depicts and how detailed it looks, with a user study behind it.
An open-source server that lets Claude drive Photoshop directly, so you can describe what you want done and watch it happen in the app.
Black Forest Labs' 4-billion-parameter model handles both text-to-image and image editing in a lighter package, which makes it worth a look if the full FLUX models have been too heavy for your setup.
Cowork lets Claude read, edit, and create files in a local folder so you can hand it real work instead of pasting things back and forth.
A custom ComfyUI node that lets you position a camera in a live Three.js viewport and spits out the prompt text your multi-angle generation needs, so you stop guessing at angle descriptions.
A Chrome extension that lets you hover over any element on any site and pull its CSS, colors, fonts, spacing, and assets straight into your workflow.
A fine-tuned open model that turns a single image into a 3D mesh, built on Hunyuan3D-2.1 and free to use commercially.
Nothing cleared the bar here yet — the standard stays high.