AI weekly, November 15–21 · Suno makes music videos and Microsoft upgrades image editing
Eight stories from the week, including music videos in Suno, two Microsoft image editors, open voice isolation and video avatars, GPT‑6.2, Gemini 4 Neon, Inkling 2 Mini, and Anthropic’s Nasdaq debut.
Suno shipped a way to make videos for songs, Microsoft opened previews of two image editors, and Inworld released weights for a speech denoiser. Of the eight stories this week, those three are the easiest to try on your own material.
Suno Visuals turns a song into a video · November 17
Suno can now make a clip or lyric video from a written scene idea or an image reference. The editor lets you extend a video on a timeline, change the text font and color, and choose export settings. Suno says Visuals is available on the web and mobile. Visual credits are purchased separately from song credits.
For a first pass, choose a chorus that listeners recognize without hearing the intro. Make a vertical clip and check whether the lyrics are readable at phone size and stay clear of the edges.
MAI Image 2.7 and Flash for precise edits · November 17
Microsoft opened public previews of two models in Foundry. MAI Image 2.7 is built for precise prompt-driven edits, while Flash is aimed at faster iterations. Microsoft also introduced Canvas for making a series of changes in one workspace.
To compare them, ask both models to change the background color of a product photo while keeping its label. Inspect the outline, small print, and reflections at full size. That is an easy way to see when Flash is enough and when the larger model is worth another pass.
LongCat-Video-Avatar 1.6 animates a character from speech · November 15
Meituan's team released an update to its open framework for audio-driven avatar video. The developers report more natural lip, face, and body movement and fewer accumulated errors across longer scenes. The repository provides code, demos, and links to the weights.
Try a short monologue with pauses and a change in tone. Compare the start and end of the same sentence. Lip sync, face consistency, and background drift are easier to spot there than in a long showcase.
Voice Isolator separates speech from the background · November 18
Inworld released the code and weights under Apache 2.0. The model takes 48 kHz mono input. The offline version is for existing recordings and the streaming version is for live audio or voice interfaces. It removes background noise while retaining the recorded voice. Restoration of missing frequencies is outside this release.
On an interview recording, listen to a quiet sentence and to a pause with background noise before and after filtering. This reveals how much noise was removed and whether quiet word endings survived.
Astra and Sol move to GPT‑6.2 · November 17
OpenAI released GPT‑6.2 Astra for complex analysis, coding, and document work, and GPT‑6.2 Sol for everyday multistep tasks. Both API models accept text and images and return text. To compare them, give both the same storyboard and ask them to find continuity errors in props and camera movement.
Gemini 4 Neon reaches the API and Antigravity · November 18
Google began rolling out Neon in the Gemini API, Google AI Studio, Gemini for business, and consumer interfaces on eligible plans. The model supports a context window of up to one million tokens and can work with several kinds of input. For a creative workflow, pair a recorded review call with the storyboard and ask for the agreed changes, with each item tied to a point in the conversation.
Inkling 2 Mini gets open weights · November 20
Thinking Machines Lab released its multimodal model on Hugging Face under Apache 2.0. It accepts text and images, returns text, and is designed for local experiments. The joint announcement reports about 26 billion physical parameters and roughly three billion active per token. Those figures describe different things. Measure actual memory use with your images and context length when choosing a setup.
Anthropic lists on Nasdaq · November 19
Anthropic shares began trading under the ticker ANT. The next day, the company announced the close of its IPO and concurrent private placement. The combined gross proceeds were approximately $119.5 billion. That figure includes proceeds from selling stockholders and is before expenses. The offering announcement did not describe any changes to Claude pricing or capabilities.
Create your own result
Choose an available model and check its price in the studio.
Sources
- Suno — Introducing Visuals
- Microsoft AI — MAI Image 2.7 and 2.7 Flash
- Meituan LongCat — LongCat-Video-Avatar 1.6
- Inworld AI — Introducing Voice Isolator
- OpenAI — Introducing GPT-6.2 Astra and Sol
- Google DeepMind — Introducing Gemini 4 Neon
- Hugging Face × Thinking Machines Lab — Inkling 2 Mini
- Anthropic — Closing of the IPO and private placement
