Weekly roundup
AI model drops: Kling Image 4.0 and Genie 4
A busy week for stills and talking heads, and Google opened Genie 4, its new world model, to Ultra subscribers. These are the four we'd actually open, and why.
Kling Image 4.0: native 8K stills and sets of ten
Kling Image 4.0 is now live in Kling Creative Studio and the API. It generates stills at native 8K, which leaves room to crop in post, and can put up to 2,000 characters of text in one image, in seven languages, with more than 70 preset fonts. One prompt can also return a set of up to 10 images that stay in the same style, which is useful for a storyboard or a product sequence, and there are more than 110 colour grading presets, or you can pull a look from a reference still. Our Kling page covers the video model, Kling 4.0, which runs takes of up to 30 seconds.
HeyGen Avatar 6: a photo that talks, with no set length
Avatar 6 turns a single photo, and optionally a short recording, into a talking video at up to 2K, with no fixed length limit. HeyGen's own examples run as continuous takes of 25 to 28 minutes, following the audio. You can write a gesture next to a line in the script and the avatar performs it on that word, and a host and guest can be generated in one frame. If you only need to change a few words in an existing recording, it regenerates those words and leaves the rest as filmed.
Genie 4: walk around a world and change it as you go
Genie 4 is now in Project Genie for eligible Google AI Ultra subscribers, with Google AI Pro on a waitlist. While you move through a world you can type a new prompt to change the weather, drop in a character or redirect the action, and for the first time the model adds real-time native audio that follows the scene, from ambience to dialogue. Google also opened it on 19 Nov through the Gemini API and Google AI Studio, where it can generate continuous 720p worlds at 24fps from a text prompt, a reference image, or both. All video and audio carries a SynthID watermark, and Google still calls it a research prototype, so camera lag and long-term scene permanence are open problems.
LongCat-Video-Avatar 1.6: open weights past the ten-minute mark
Meituan released an upgrade to its open-weight, audio-driven avatar model, under an MIT licence tagged Commercial OK on our Model drops line. Compared with 1.5, it aims for steadier lip sync and more natural face and body motion, and Meituan says it can keep identity and quality in clips that run past 10 minutes. If you want to run an avatar model yourself rather than through a hosted app, this is the one to try.
Also dropped
- Gemini 4 Neon Google, 18 Nov 2026
Google's mid-tier Gemini 4 model for complex coding, analysis and long agent work, in the Gemini API, AI Studio, Gemini for business and the Gemini app for Pro and Ultra subscribers. - MAI Image 2.7 Microsoft AI, 17 Nov 2026
Microsoft's new image generation and editing model, with a cheaper MAI Image 2.7 Flash, in public preview on Microsoft Foundry and in MAI Labs. - GPT-6.2 Astra and Sol OpenAI, 17 Nov 2026
OpenAI's new pair of language models, Astra for the hardest projects and Sol for everyday work, in ChatGPT Work, Codex and the API. - Hy4 flash Tencent, 16 Nov 2026
Open-weight multimodal language model from Tencent's Hy family, a mixture-of-experts model with 12B active parameters that reads text and images with a 1M-token context. - Qwen4-Max Alibaba, 16 Nov 2026
General release of the flagship Qwen model in Alibaba Cloud Model Studio, reading text, images and video with deep reasoning, tool calling and web search.
Related on AI Field Notes
- Kling: our model notes Model notes
- Film news, November 2026 News month
Every claim here comes from the maker's own page, linked above. Licence tags are what the maker's licence or terms said on the day we filed, not legal advice, so check the linked terms before client work. The full day-by-day list is on Model drops, and older roundups are in the weekly archive.