A curated list of AI generators
Text, image, video, voice, music, code and 3D tools, with the latest models. 98 tools in 9 categories.
What's new
The releases and shake-ups that changed this list in the past year, newest first.
OpenAI's Sora API shuts down
The Sora web and app experiences closed on April 26. With the API gone too, Seedance, Veo, Kling, Hailuo, and Runway are the video models to look at.
Suno ships v6
A new model line in three sizes (v6, v6-wild and v6-mini), built jointly with its label partners.
Midjourney V8.2 and Seedream 5.0 Pro
Midjourney adds an editing model and better personalization. ByteDance's Seedream 5.0 Pro adds precision editing and native multilingual text.
GPT-6 Astra arrives
OpenAI's next flagship follows the GPT-5.6 line that launched in July.
Google previews Lyria 3.5
Clip and Pro versions of Google DeepMind's music model enter preview.
Claude Fable 5.1 and Mythos 5.1
Anthropic's top-tier models get their 5.1 update. Fable is generally available; Mythos stays with approved organizations.
Show earlier changesHide earlier changes
A crowded month for open weights
Qwen3.8, GLM-5.3 and a vision update to DeepSeek V4 Flash all land, keeping open models close behind the closed leaders.
GPT-5.6 launches publicly
Sol, Terra and Luna cover the range from top capability to low cost.
Claude Science for researchers
Anthropic's research workbench arrives, alongside Google's Gemini for Science suite from I/O.
Stable Audio 3
Stability AI's latest text-to-audio model, with an API and builder-friendly terms.
DeepSeek V4 under the MIT license
Open weights with a 1M-token context, in Pro and Flash sizes.
GPT Image 2
OpenAI's newest image model, with stronger text rendering and editing, replaces DALL-E in ChatGPT.
Kling 3.0 and Seedance 2.0
Native audio, multi-shot storyboarding and multimodal references move AI video closer to scenes than clips.
Project Genie opens to US Ultra subscribers
A web prototype built on Genie 3 turns a prompt into an explorable world.
The music labels settle
Universal settled with Udio, and Warner settled with both Suno and Udio. Both services are moving to licensed catalogs.
Six shifts to know about
What changed the way people choose between tools.
Video learned to talk
Veo 3.1 and Seedance 2.0 generate sound and picture together, and Kling 3.0 plans multi-shot scenes. OpenAI's Sora is no longer available.
Open weights caught up
DeepSeek V4, Kimi, GLM and Qwen sit within a few points of the closed leaders on public indexes, mostly under MIT or Apache 2.0 licenses.
Music went licensed
Warner settled with Suno and Udio, Universal settled with Udio, and both are rebuilding around licensed catalogs. Terms are still moving.
Coding became agentic
Claude Code, Codex and Cursor's parallel agents handle whole tasks across many files. Surveys put daily AI use among professional developers well above half.
No single best image model
GPT Image 2 for text and editing, Nano Banana Pro for dense graphics, Midjourney for look, Flux when you want open weights.
Science tools turned agentic
Co-Scientist, AlphaEvolve and Claude Science propose hypotheses and search for algorithms instead of only predicting structures.
Image generators
Picking an image model is now a routing decision: one is best at text, another at look, another at editing.
General tools (text to image)
OpenAI's image model inside ChatGPT and the API, and the successor to DALL-E. Follows long prompts closely, renders readable text and edits conversationally.
GPT Image 2Apr 2026
Google's image models in the Gemini app, AI Studio and the API. Nano Banana Pro handles dense instructions and information-rich scenes; Nano Banana 2 is the faster tier.
Nano Banana Pro and 22026
Still the one for aesthetics and art direction. V8 reads prompts more literally than V7, and V8.2 adds a new editing model and stronger personalization.
V8.2Sep 2026
Black Forest Labs' FLUX.2 family: high quality, strong prompt adherence and text rendering. The smaller Klein models are open-weight and run locally; the pro models are API-only.
FLUX.22026
ByteDance's image family, used in Dreamina. Seedream 5.0 Pro adds interactive precision editing, position-aware edits, native multilingual text and denser information graphics.
Seedream 5.0 ProSep 2026
Typography first: posters, logos and anything with words in it. Has a free plan.
Designer-minded generator with native.
Built into Photoshop for generative fill and expansion, and trained with commercial use in mind, which matters for brand and agency work.
SDXL, SD3 and the community ecosystem of fine-tunes, LoRAs and ComfyUI workflows. Still the base for custom, self-hosted pipelines.
Alibaba's image model family, with open-weight releases on Hugging Face.
A platform for designers and game artists that bundles several models with an AI canvas for editing.
A free, easy way to start. The quality ceiling is lower than the paid leaders, but it is enough to learn how prompting feels.
Specific applications
Personalized art, powered by AI. Highlights: photo to art, vectorize to pdf/svg.
Photo to art, supports 80+ art styles.
Background removal and an AI photo editor.
Coloring page generator.
Text generators
Chatbots, reasoning models and writing tools. Frontier releases now land every few weeks, and open-weight models trail the leaders by only a few points.
General tools (LLMs, chatbots, question answering, summarization, writing)
OpenAI's assistant, with fast, thinking and pro model tiers plus image generation, voice and agent features. The GPT-5.6 line (Sol, Terra, Luna) launched in July.
GPT-6 AstraSep 2026
From Anthropic. Strong at writing, long-document analysis and coding, with Claude Code and Claude Science for work in the terminal and the lab. Models run from Haiku 5.5 through Opus 5.5, with Fable 5.1 on the new top tier.
Fable 5.1Sep 2026
Google DeepMind's multimodal assistant, built into Search and Workspace. The 3.x line comes in Pro and fast Flash tiers, with Gemini Omni for generating content across media.
Gemini 3.8 FlashSep 2026
xAI's assistant, with live access to X. Also generates images and video through Grok Imagine.
Grok 4.x2026
MIT-licensed open weights with a 1M-token context, strong at coding and agent work at a low price. V4 comes in Pro and Flash sizes.
V4Apr 2026
Alibaba's model family, with open weights from phone-sized to very large, mostly under Apache 2.0. Also available through Hugging Face Chat.
Qwen3.8Aug 2026
Moonshot AI's long-context, agent-focused models. K2.6 led open-weight rankings in spring; K3 targets long multimodal agent tasks and needs serious hardware to self-host.
Kimi K32026
Z.ai's models, with MIT-licensed weights aimed at long-horizon coding and reasoning.
GLM-5.3Aug 2026
A European provider with Apache 2.0 open models (Large 3, Small 4) and the Le Chat assistant. A good fit when data residency matters.
Meta's Llama 4 family (Scout, Maverick). The deepest Western tooling ecosystem; read the community license first, especially for use in the EU. Also on Hugging Face Chat.
Google's small open models, built to run on a single consumer GPU. Gemma 4 is released under Apache 2.0.
Gemma 42026
OpenAI's open-weight reasoning models (120B and 20B) under Apache 2.0.
Search
AI-powered search: concise answers with sources, combining an LLM with web search.
Writing assistants
Search, generate, analyze and chat right inside Notion.
AI writing assistant.
For students: helps with reading, writing and referencing.
For sales and marketing: sales copy, digital ads, blog posts, email and product descriptions.
For sales and marketing: AI content writing and marketing copy.
Notes and research
From Google. An AI note-taking and research assistant that works from your own sources, and can turn them into a podcast-style audio overview in one click.
Video generators
The biggest shift this year: video models now generate sound with the picture, and OpenAI's Sora is gone.
General tools (text to video, image to video)
ByteDance's model that generates audio and video together from text, image, audio and video references. It led blind image-to-video rankings in June. Seedance 2.5, aimed at longer, reference-heavy production.
Seedance 2.0Feb 2026
Google's flagship video model, with synchronized audio, up to 4K output and vertical video. Use it in the Gemini app, in Flow or through Vertex AI. A cheaper Lite tier is available for silent clips.
Veo 3.12026
Video model from Chinese company Kuaishou. Strong human motion, native 4K at 60 fps, 15-second clips, multi-shot storyboarding and lip sync. A free daily credit allowance is included.
Kling 3.0Feb 2026
The closest thing to an editor among the generators: motion brush, camera control, video-to-video editing (Aleph) and 4K ProRes export.
Gen-4.52026
Fast generation and strong camera motion. Ray 3 can anchor a clip to a first and last frame.
Ray 32026
Special effects and playful transformations. Pikaframes interpolates between a start and end frame.
MiniMax's video model, known for fast image-to-video.
Alibaba's open video model, and the main option for running video generation yourself. Wan 2.6 is Apache 2.0.
Wan 2.62026
xAI's text and image to video, priced aggressively.
Video for brands and agencies that need commercially safe output.
- SoraSunset
OpenAI shut down the Sora web and app experiences on April 26, 2026, and the API on September 24, 2026. Listed so older links and recommendations make sense.
Specific applications
AI videos with virtual avatars and voiceovers.
Create and translate videos with AI avatars.
Prompt to ready-to-post social video.
Faceless short videos on autopilot, with auto-posting to TikTok and YouTube.
Voice generators
Speech now comes with emotion and multi-speaker dialogue, and voice agents are a product category of their own.
General tools
The most realistic speech generation: Eleven v3 with multi-speaker dialogue, voice cloning, dubbing, sound effects and music, plus ElevenAgents for building voice agents.
Eleven v3Feb 2026
Low-latency speech models (Sonic) built for real-time voice agents.
Sonic 3.6Aug 2026
Text to speech and speech to text APIs.
Human-level speech recognition (speech to text) from OpenAI. Open source, and you can try it on Hugging Face.
Music generators
AI music is becoming a licensed business. Check each service's current terms before you release anything.
The default for full songs, with the most natural-sounding vocals. Has signed licensing deals with Warner Music, BMG and Believe/TuneCore; v6 is a new model line built with those partners.
v6Sep 2026
Finer control over the sound and cleaner long-form output, but in transition. After settling with Universal and Warner it is rebuilding as a licensed platform, and downloads have been restricted, so check before relying on it.
Google DeepMind's music model. Lyria 3.5, in Clip and Pro versions, is in preview.
Lyria 3.5Sep 2026
Stability AI's text-to-audio model, with an API and terms that work for builders adding music to an app or game.
Stable Audio 3May 2026
A text-to-music model from ElevenLabs that can switch genres mid-track.
Code generators
Autocomplete is no longer the main event. The leading tools are agents that plan, edit many files, run tests and work in parallel.
Coding agents and editors
Anthropic's agent for engineers, in the terminal, on the web, desktop and mobile. Favored for multi-file refactors and long, hard tasks.
OpenAI's coding agent across CLI, IDE extension and cloud app. Good for delegating backlog tasks, and included with ChatGPT plans.
An AI-native editor (a VS Code fork) with parallel agents in separate git worktrees and fast visual diffs.
The most widely adopted assistant. Now includes agent mode, cloud agents, code review and a CLI inside VS Code and other editors.
Google's open-source terminal agent, with a generous free tier.
Cognition's autonomous software engineer, run as a background agent.
Open-source coding agent for VS Code that works with the model of your choice.
Open-source terminal agent that works with any model provider, including local models.
The original terminal pair programmer, open source and git-native.
A browser IDE with an agent that builds and deploys apps from a prompt.
The models behind most of these tools are the general ones in Text generators: Claude, GPT and Gemini, plus open models such as DeepSeek and Qwen.
3D generators
Image-to-3D has matured into production pipelines with PBR textures and rigging, and world models now generate whole explorable scenes.
Image or text to 3D object
Fast, clean text-to-3D and image-to-3D, with auto-rigging.
Text, image and multi-view to 3D, with PBR texturing, rigging and slicer and engine plugins.
Meshy 6
Text, image and sketch to 3D with PBR textures. The hosted version has a free daily quota.
Top-tier geometry and PBR, quad topology and multi-image input.
Gen-2
Procedural plus AI generation for game-ready assets.
Text to 3D.
3D worlds and scenes
Fei-Fei Li's world model. Generates persistent, explorable 3D worlds (Gaussian splats plus meshes) from text, images, video or a rough 3D layout. Has a Chisel editor and a developer API.
Marble 1.1 / 1.1 Plus
Google DeepMind's prototype for generating interactive worlds from text or images, built on the Genie 3 world model. Limited to US AI Ultra subscribers, with 60-second sessions at 720p.
Genie 3Jan 2026
Reconstructs 3D objects and people from a single image.
Open source and local
Tencent's open model with a full PBR texture pipeline.
Image to 3D with PBR output. Runs in ComfyUI.
AI tutors
The best tutors are built to ask you questions rather than hand over answers.
Khan Academy's K-12 tutor, designed to coach rather than answer. Free for teachers and about $4 a month for families.
Turns ChatGPT into a Socratic coach that asks questions and breaks problems into steps. Available on every plan, including free.
Guides you with questions instead of answers, and works inside Claude Projects. Part of Claude for Education.
Built on LearnLM: step-by-step guidance with images, diagrams, videos, quizzes and flashcards.
Magic Notes turns your notes into study sets, and Q-Chat quizzes you conversationally.
More than 80 lesson-planning tools for teachers, many in the free tier.
A language learning app that gets you speaking with AI.
Practice English conversations with AI.
AI for science
Science tools have moved from predicting structures to proposing hypotheses and searching for new algorithms.
Biology
Predicts a protein's 3D structure from its amino acid sequence. Won a Nobel Prize and has been cited tens of thousands of times.
DeepMind's model for understanding the effect of genomic variants.
AlphaGenome2026
Research agents
Google's multi-agent system, in which agents generate, critique and rank hypotheses like a research team. A Nature paper in May 2026 reports its results.
A Gemini-powered coding agent that evolves new algorithms against automated evaluators. It has produced new math results and infrastructure speedups.
Google's suite announced at I/O that brings Co-Scientist, AlphaEvolve and its other science systems under one brand.
Anthropic's research workbench for scientists, run on the lab's own machines. It can spin up sub-agents and split a research task between them.
Math and chip design
Advanced reasoning in mathematics; solved International Mathematical Olympiad problems at silver-medal level. Not released.
A chip design assistant. Not released.
No tools match that search.