AI Notes September, October, November 2025

AI Tools for Creatives – September, October and November 2025

August–November 2025 marked a shift from “wow” demos to always‑on AI agents quietly doing work in the background. GPT‑5.1, OpenAI’s Atlas browser, proactive features like Pulse, and new security agents like Aardvark pushed assistants closer to autonomous coworkers, while Anthropic’s record‑setting author payout and Udio’s peace deal with Universal Music signaled that copyright and compensation are finally being priced into the system.

General AI tools

GPT-5 / GPT-5.1 — (Major Launch, Updated) OpenAI’s GPT‑5 launch in August unified GPT and o‑series with routed “Thinking” vs “Instant” modes for deeper reasoning and faster responses, then was quickly followed by GPT‑5.1 which tightened multi‑step reasoning, sped up responses, and expanded tone/personality controls across ChatGPT, Copilot, and API products. GPT‑5.1 now underpins Pro and enterprise tiers, with extended prompt caching and better tool use for multi‑app workflows.

Claude Opus 4.5 — (Updated) Claude Opus moved from “just” a reasoning beast to a more introspective, agent‑ready model, with Anthropic releasing research showing limited but genuine introspection and deploying Opus across coding and business workflows. In parallel, Anthropic agreed to pay at least $1.5B to authors and delete unlicensed datasets, likely the largest copyright settlement yet and a blueprint for future training‑data deals.

Perplexity — (Updated) Perplexity Comet, its AI browser, is now being mass‑distributed through PayPal and Venmo, bundling a year of Perplexity Pro for millions of users to accelerate adoption in the browser wars. A new Email Assistant for Max subscribers lives directly in Gmail/Outlook, drafting replies, organizing messages, and scheduling meetings while mirroring your tone without training on private email, pushing Perplexity deeper into knowledge‑worker workflows.

Gemini 2.5 / 3 — (Updated) Gemini 2.5 Deep Think’s high‑end reasoning is now complemented by Gemini 3 models rolling into Search and “AI Mode,” which expands to multiple new languages and positions AI‑led results as the default experience for hundreds of millions of users. On the enterprise side, Vertex AI’s agent tooling continues to mature, bringing guard‑railed agents into production stacks.

Microsoft Copilot — (Updated) Copilot’s GPT‑5 upgrade, Copilot 3D, and deeper Windows 11 integration continued to roll out, with new experiments like a “Share with Copilot” taskbar button that beams any current window into Copilot Vision for on‑screen assistance. Across Office and Windows, Microsoft keeps adding more “agent hooks” rather than new standalone apps.

ChatGPT Pulse — (New) Pulse is ChatGPT’s new proactive research mode for Pro users: it works overnight, then delivers a handful of visual “cards” in the morning summarizing news, priorities, calendar items, and custom topics it learned you care about. The key design choice is that Pulse deliberately stops after a few cards with “that’s it for today,” framing itself as an assistant, not an addictive feed.

ChatGPT Atlas — (New) Atlas is OpenAI’s macOS browser with ChatGPT built‑in and an agent mode that can read pages, fill forms, book reservations, and remember your browsing context if you opt in. Browser memories let you ask things like “summarise the jobs I looked at last week,” while the agent can research, plan, and act directly through the browsing context, raising both productivity potential and fresh security concerns.

Creative AI tools

Adobe Acrobat Studio — (Updated) Acrobat Studio’s PDF Spaces concept matured as a backbone for “document as knowledge base,” with agents able to reason across multi‑file collections and generate conversational answers, structured outputs, or workflows from PDFs. Adobe MAX 2025 then spotlighted Spaces alongside Firefly‑driven design features and integrations into Creative Cloud and Express, reinforcing Adobe’s strategy of AI as infrastructure rather than novelty.

Google Gemini 2.5 Flash Image (“nano‑banana”) — (Updated) The infamous “nano‑banana” image model has now been rolled formally into Google tooling as Gemini 2.5 Flash Image, with creators using it for layout‑aware, text‑accurate, character‑consistent images. New developer tooling from Google and partners like Leonardo makes it easier to prototype with Flash Image in real workflows, from product shots to concept art.

Leonardo AI (Lucid Origin, FLUX Kontext) — (Updated) Leonardo’s Lucid Origin model continues to deliver vibrant, Full‑HD renders for branding and graphic design, while FLUX .1 Kontext now supports up to four reference images blended via inline text instructions, allowing very tight creative direction from a small style library. Combined with real‑time canvas and 3D mesh export, Leonardo is leaning into production pipelines rather than one‑off generations.

Recraft — (New) Recraft doubled down on brand‑consistent assets: you define visual “DNA” via style presets and references, and the system enforces that across icons, logos, ads, and mockups, which is particularly useful for small teams without in‑house brand designers.

Midjourney HD Video — (New) Midjourney's entry into video generation with dynamic sequences, adjustable camera angles, and smooth transitions from text prompts.

Higgsfield — (Updated) Added Upscale feature for boosting photo resolution up to 16x and video up to 4K, plus Higgsfield Assist with GPT-5 integration for creative prompting.

Ideogram Character Reference — (New) Addresses character consistency in AI images using single reference photos to maintain core features across different scenes and styles.

Adobe Firefly — Structure Reference continues to allow layout locking while varying creative content, plus collaborative Boards feature.

Adobe Illustrator — Generative Shape Fill tool remains a key feature for vector art creation.

Midjourney V7 — Enhanced video generation modes, improved logo creation, and new editor capabilities.

Story Diffusion — Maintains comic character consistency across panels.

DALL·E 3 — Inpainting and regional editing directly within ChatGPT.

Khroma — AI color scheme generator that learns user preferences.

AutoDraw — Quick sketch helper for wireframes.

Fontjoy — Font pairing generator.

Botika — AI fashion models for Shopify product images.

Magnific — AI upscaler with prompt-guided detail enhancement.

Presti — Realistic furniture placement in AI-generated rooms.

IC Light V2 — Text-guided portrait relighting tool.

Krea 3D Objects — Image to textured 3D mesh generation.

Flora — Team collaboration tool chaining text, images, and videos.

Stable Virtual Camera — Generates explorable 3D scenes from photographs.

Wizard of Oz @ The Sphere (Gemini + Veo) — (New) A remastered Wizard of Oz at Las Vegas’ Sphere used Gemini, Veo, and Imagen to extend scenes and rebuild compositions for 360° projection, with more than 90% of the film touched by AI to preserve original performances while fitting the new format. It’s an early proof‑of‑concept for using generative AI to retrofit legacy IP into radically different experiences without fully re‑animating from scratch.

Other creative standouts from August are still very relevant: Ideogram 3.0 for text‑in‑image design, Midjourney HD Video and V7 for cinematic sequences, Leonardo’s 3D exports, DALL·E 3 for in‑ChatGPT editing, and specialised tools like Botika, Magnific, IC Light V2, and Krea 3D Objects for fashion, upscaling, lighting, and geometry workflows.

Video AI tools

Google Vids — (Updated) Vids is now part of the Workspace story: image‑to‑video, AI avatars, filler‑word removal, and script‑to‑timeline generation live next to Docs, Sheets, and Slides, making “deck + video” output feel like a single workflow rather than a handoff.

Veo 3.1 — (Updated) Veo 3’s image‑to‑video mode quietly became mainstream via Google Photos’ “Photo to Video” feature, generating four‑second clips from still images for 1.5B users, with AI Pro/Ultra tiers getting higher limits and quality. It also integrates into creator stacks like Leonardo and YouTube’s Shorts tooling, placing Veo at the core of Google’s consumer video AI ecosystem.

Runway Aleph — (Updated) Runway’s Aleph engine continues to serve as a flexible front‑end for multiple video models, including Veo 3, wrapped in a text‑first editor that supports more granular, timeline‑aware controls.

MovieFlo AI — (New) MovieFlo’s scene‑based workflow and adoption of the best current models (Veo, Kling, Runway, Dream Machine, Nova, etc.) positions it as a “director’s console” for AI‑powered pre‑viz and finished sequences, especially attractive to film‑adjacent teams with existing pipelines.

Sora Cameos — (Updated) Sora now supports “cameos,” letting you bring recurring characters, pets, or props into new shots while maintaining appearance and motion style, sitting alongside the anticipated Sora 2 improvements in motion and synchronized audio. The Critterz project uses Sora‑type tooling to test whether a nine‑month, sub‑$30M animated feature is viable, eyeing a Cannes 2026 debut.

Kling — Fast and controllable 1080p generative video tool for social media content.

Scenario — Direct video generation control through frame sketching.

Seedance 1.0 — Narratively consistent video creation with multi-shot character cohesion.

Mirage (Decart) — Real-time world transformation and style transfer for video streams.

Moonvalley Marey — Hybrid 2D-to-3D filmmaking AI platform with camera motion control.

Dream Machine (Ray 2) — Cinematic AI-generated clips with realistic physics.

Amazon Nova Reel — Budget-friendly text-to-video generation.

Pika Pikas — AI actor and background swaps via prompts.

Gemini Video Generator — AI-produced animated scenes.

Descript — Video/audio editor with overdub and eye contact correction.

Viggle — Motion transfer and greenscreen-ready character creation.

Synthesia — Avatar video creation with multilingual voiceovers.

Facefusion — Open-source GPU-based face-swapping tool.

Deep Live Cam — Real-time deepfake streaming for VTubers.

Revid.ai — Auto video summarization for social shorts.

Riverside — Lossless recording with text-based editing.

Flux — Fast 1080p video generator with style control.

HeyGen — AI face swaps with multilingual voice control.

Arcads — Automated ad video creation platform.

Filmora AI — Video editing with auto cut, style transfer, and filler removal.

Keytake — Converts documents and URLs into explainer videos.

The August set of video tools—Kling, Scenario, Seedance, Mirage, Dream Machine (Ray 2), Nova Reel, Pika, Gemini Video, Descript, Viggle, Synthesia, Facefusion, Deep Live Cam, Revid, Riverside, Flux, HeyGen, Arcads, Filmora AI, and Keytake—remain very much alive and continue to fill specific niches from deepfakes and avatars to podcast editing and explainer‑video generation.

Music & audio AI tools

ElevenLabs Music — (Updated) ElevenLabs’ expansion into music coincided with its CEO publicly arguing that core audio models will be commoditised soon, shifting competition toward integration and UX rather than raw model quality. The company is exploring multimodal setups that blend voice, music, and LLMs along lines similar to Google’s Veo 3 pipeline.

Udio x Universal Music Group — (New) Universal Music Group dropped its lawsuit and instead signed a licensing partnership with Udio to launch a fully licensed AI music platform in 2026, trained on authorised content and fenced with fingerprinting and filtering to keep infringing material contained. Udio’s existing consumer product will remain in a more restricted, monitored environment, but the deal shows how quickly “AI vs labels” battles can pivot into commercial alliances.

Suno V4.5 — Fast 2-track mixing with improved vocals.

ElevenLabs — Text-to-speech and sound effects, now with music generation capabilities.

Play HT — High-quality, low-latency voice API.

Artlist AI Voiceover — Multilingual voiceovers with Adobe Premiere support.

Hume Octave — Emotion-aware text-to-speech with controllable tone.

SoundHound Chat — Car assistant AI for voice commands and ordering.

ElevenLabs Bark — AI speech technology for pet-tech toys.

Amazon Nova Sonic — Expressive voice AI powered by AWS Bedrock.

Synthflow — AI tool for automating meeting bookings and CRM updates.

Shamaze — Generates bedtime stories in personalized AI voice.

The rest of the August list—Suno V4.5, Udio, PlayHT, Artlist AI, Hume Octave, SoundHound Chat, Amazon Nova Sonic, and story‑adjacent tools like Shamaze—still offers a complete audio stack from emotional TTS to in‑car assistants.

UI design AI tools

The core UI tools from August remain the go‑tos:

• Figma “First Draft” — AI layouts and content for early explorations.
• Uizard, Galileo, Dora — text‑to‑screen and no‑code 3D site workflows.
• Webflow AI Site Builder, — brief‑to‑site pipelines.
• Attention Insight, Operative.sh — predictive gaze and UX testing.
• AI CSS Animations, Same.dev, Firebase Studio — cloned front‑ends, animations, and back‑end scaffolding.

Marketing AI tools

• Pencil — now on Google Cloud Marketplace for easier enterprise procurement.
• Jasper, Virallyst — content, influencer, and social optimisation.
• Warmy — domain warm‑up and deliverability.
• Outreach agents like AiSDR and Reachy.
• Landing‑page tools like Keak.
• Happenstance — Plain language AI network search.

Ragebait‑driven growth from products like Cluely and how controversial messaging plus AI‑assisted content can drive awareness, but those are more cautionary case studies than tools to recommend.

No Code Builders (Vibe coding)

Lovable — (Major Update, Still Hot) Lovable’s August story—projected $1B ARR, $8M ARR monthly growth, and vibe‑coding agents—carried into autumn, with more dev teams moving from traditional IDE workflows to natural‑language, agent‑assisted coding.

Replit Agent — (Updated) The new Replit Agent now supports autonomous workflows that can generate, run, and debug applications in a browser, along with tooling to “generate agents” that operate on top of Replit projects themselves.

Warp Code — (New) Warp’s agentic terminal became a recurring mention: top‑tier SWE‑bench scores, MCP/CLI integrations, and a UX tuned specifically for tight human‑in‑the‑loop agent work rather than black‑box automation.

Junie by JetBrains — (New) Junie is JetBrains’ coding agent baked into its IDEs, providing stepwise, transparent actions and summaries, and fitting directly into existing Git workflows instead of spinning up a parallel coding environment.

V0 by Vercel, GitHub Copilot, and Emergent from August remain staples for front‑end scaffolding and mobile app generation, now often orchestrated by agent frameworks for multi‑tool workflows.

Local Models and Tools to Run Them

Tools

Ollama — Open-source tool for running LLMs locally with command-line interface and API.

Chatbox — Cross-platform AI client for Windows, macOS, Linux, Android, iOS, and web.

Open Web UI — Self-hosted AI interface supporting Ollama and OpenAI-compatible APIs.

Comfy UI — Node-based interface for generative AI with visual workflow creation.

Automatic 1111 — Browser-based interface for Stable Diffusion with advanced tools.

Models

OpenAI GPT-OSS — (New) Open-source models (120B & 20B), optimized for reasoning and coding, Apache 2.0 licensed.

DeepSeek R1-Omni — 671B open-weights model with 200k context, free for research.

Llama 3 (8B/70B) — Llama 3.1 with 405B-parameter model, 128K token context, multilingual coding. Llama 3.2 brings vision-capable and edge-friendly models.

LG EXAONE Deep 32B — Laptop-friendly model scoring near GPT-4 on STEM tasks.

Ethical Models

FLite — Open-source 10B-parameter diffusion model trained on 80M licensed images from Freepik.

Bria AI — Compact open-source text-to-image model (4B parameters) built on licensed data.

Blunge — Ethical AI image generation protecting artists' rights through manual ownership checks.

Other AI tools

OpenAI Aardvark — (New) GPT‑5‑powered, agentic security researcher that continuously scans codebases, identifies and validates vulnerabilities, then proposes fixes with annotated explanations. Benchmarks report high detection of known and synthetic bugs on curated repos, and it’s already running across OpenAI’s own codebases in private beta.

Microsoft MAI-Voice-1 & MAI-1-preview — (New) Microsoft's first in-house AI models, with MAI-Voice-1 generating one minute of audio in under one second.

OpenAI GPT Realtime — (New) Real-time API for building voice agents with 82.8% audio reasoning accuracy.

Anthropic Claude for Chrome — (New) Chrome extension allowing Claude to handle browser tasks like cart management and form filling.

Gradio — Open-source Python library for building interactive ML web apps.

Wide Research by Manus — Tool for handling multiple research tasks concurrently.

Harvey — AI bot for contract review and legal due diligence.

CopyCat — Low-code browser automation from natural language instructions.

Exa Search — Hybrid semantic+keyword search API for docs and e-commerce.

Lambda Inference API — Pay-per-token gateway to major frontier models.

Zapier MCP — One prompt triggers 8,000+ SaaS actions for agents.

Kimi K2 — Chinese open-weight 1T-parameter LLM outperforming GPT-4 in coding/math.

Payman — Agent-driven hiring with secure payment.

Pinokio — One-click local deployment of AI apps.

Cursor v1.3 — Shared terminal and faster context-aware coding chat.

Browser Use — Library for headless browser automation in agents.

Databutton MCP — Drag-and-drop AI workflow builder.

Documenso — Open-source DocuSign alternative.

Terra Security — AI-driven penetration testing platform.

Agent Simulate — Synthetic user load testing for UX research.

From August, a lot of the “Other AI tools”—Microsoft MAI‑Voice‑1, OpenAI GPT Realtime, Claude for Chrome, Wide Research, Harvey, CopyCat, Exa Search, Lambda Inference API, Zapier MCP, Kimi K2, Pinokio, Cursor v1.3, Databutton MCP, Documenso, Terra Security, Agent Simulate, Payman—remain solid bets and are consistently referenced as part of the “agent plus infra” layer.

Education AI tools

• Google Gemini Guided Learning — interactive tutoring with visuals, flashcards, and step‑wise help.
• OpenAI Study Mode — prompts users to reason instead of handing over answers.
• Claude for Education and OpenAI Academy — large‑context tutors and structured courses.
• OpenAI Academy — Courses on prompt engineering and AI safety.
• AI Tutor by Roadmap.sh — Interactive study tool following coding/learning roadmaps.
• TurboLearn — Note-taking, flashcards, and quizzes from various media.
• NotebookLM Audio/Video Overviews — Podcast-style or visual learning summaries powered by AI.
• Nvidia free AI courses — Hands-on training for AI and ML fundamentals.
• Globe Explorer — AI-based interactive knowledge maps.
• Class Central — Indexed catalogue of online AI courses.
• University of Illinois AI — MBA-level AI specialization online.
• Microsoft Generative AI Beginner — Twelve-lesson introductory curriculum.
• Maven AI Bootcamps — Cohort courses on safety, prototyping, product.

Worth checking out

• Halo X — AI glasses with on‑device recording and analysis, still drawing heavy privacy scrutiny.
• GeoSpy — location inference from images, being tested by law enforcement with obvious surveillance implications.
• Figure 02 Helix — (New) Humanoid robot with AI model for complex tasks like folding laundry autonomously.
• Devin 2.0 — Autonomous developer agent.
• Convergence Parallel — Multi-agent orchestration framework.
• Mistral OCR API — High-speed multilingual OCR API.
• NotebookLM "Discover Sources" — Curated research companion.
• Keytake — Converts documents into branded explainer videos.
• AI Creativity Summit LA — A cross‑disciplinary, hands‑on summit in Los Angeles on November 19 that brought together artists, brands, studios, and technologists around themes like “Wild Imagination vs Replication” and “The Studio Revolution,” underlining how fast creative pipelines are changing.
• Chroma Awards — A new AI film/music/games competition offering $175K in cash prizes and $1M in credits, meant as an “Olympics of AI creativity” to surface the next wave of AI‑native storytellers.

Think pieces and resources

• Google Prompt Engineering Guide — Best practices for prompt design.
• State of AI 2025 — Concise trend graphs and insights.
• Agent Survey 2025 — 264-page review on autonomous agents.
• World Economic Forum Future Jobs Report — Skills and wage impact analysis.
• Hugging Face SmolAgents Course — Build lightweight AI agents.
• 601 AI Income Ideas — Monetization guide for AI applications.
• IBM Building AI-Powered Chatbots — Vendor-neutral course.
• Elements of AI (Stanford/Harvard) — Free fundamentals of AI and ethics.
• OpenAI hallucination incentives — OpenAI’s new paper argues hallucinations are partly a function of benchmarks that reward confident guessing, and proposes rewarding “I don’t know” to train more honest models.
• AI bubble vs infrastructure — Multiple essays and interviews debate whether current AI valuations and private IPOs constitute a bubble or whether we’re simply repricing foundational infrastructure, especially given huge projected annual AI capex by major players.
• Dead internet theory, bots, and authenticity — Sam Altman’s “everything feels fake” comments and bot‑heavy social stats sparked serious conversation about how LLM tone bleeds into human writing, and what that means for trust and discovery online.

September–November 2025 trends and market highlights

Agentic browsers & proactive assistants — Atlas and Pulse together show OpenAI’s strategy: AI that doesn’t just answer but anticipates and acts—researching overnight, then using browser‑level access to help you execute tasks. Google’s AI Mode and Perplexity’s Comet push the same direction: search and browsing as a conversation with an embedded agent rather than a list of links.

Security, safety, and copyright get priced in — Anthropic’s $1.5B author settlement, Udio’s partnership with Universal, YouTube’s likeness‑detection tools, and state‑level regulation show that copyright, consent, and catastrophic‑risk governance are no longer theoretical debates—they’re line items on P&Ls and product roadmaps. OpenAI’s Aardvark also reframes AI as a defender, not just an attack surface.

The open‑source & infra arms race — Massive private rounds into open‑source‑oriented startups plus national‑scale investments in data centres highlight that “who owns the stack” is still wide open, even as GPU and data‑center buildouts skyrocket. Open‑weights like DeepSeek and GPT‑OSS give devs more autonomy in that landscape.

From tools to orchestration — New frameworks (Microsoft Agent Framework, Salesforce Agentforce, orchestration/observability stacks) signal that the challenge is no longer “can we build an agent?” but “can we coordinate and monitor dozens of them safely in production?”. QA, logging, and monitoring tools are quietly becoming the most important pieces of the stack.

Creative AI grows up — From Wizard of Oz at the Sphere to AI Creativity Summit LA and the Chroma Awards, AI in film, music, and games is moving from experimental shorts to flagship stage shows, studio projects, and structured competitions. The fight is less about whether AI belongs in the pipeline and more about credit, consent, and keeping “wild imagination” from collapsing into “just automating the old playbook”.

Dead internet & authenticity crisis — Altman’s “dead internet” rant, Meta’s safety missteps, and the spread of ragebait‑driven, AI‑assisted marketing underline a cultural shift: AI is now as much about what we see and believe as what we build. Tools for provenance, moderation, and reputation will be just as important as the next model release.

If you spot any missing links, please DM or comment!