AI News: Google’s Image Model Becomes More Affordable, Voices and Video Gain Strength
A quiet update from Nano Banana, a new voice at ElevenLabs, longer videos in Runway — and Claude becoming a single app.
Figure 01.0 — AI News
First, a quick glossary. An image model is the AI that creates an image from a text command (a prompt). The API is the backdoor through which developers integrate such a model into their own software; you pay per image created. Reference images are photos you provide so the AI knows what someone or something should look like.
Google Nano Banana 2.1: Quietly Released, Half the Price
Google released Nano Banana 2.1 on 6 October. It is an update of Nano Banana 2, also known as Gemini 3.1 Flash Image. The model first appeared in Flow’s model selector, disappeared, and then reappeared a day later. Without much fanfare. letsdatascience
What’s improved? Text in images, prompt adherence, and maintaining consistency of people across multiple steps. It works with up to 14 reference images at once: up to four people and ten objects that must remain the same. Where necessary, it checks Google Search before drawing real subjects. arynewsarynews
And the price: Google approximately halves the price per image for developers. However, there’s a catch: the price for input (text, image, and video you send along) actually increases. Those who send many reference images gain less than the headline suggests. arynewsarynews
How much better is it really? Google cites its own figures. According to ARY News, 2.1 scores higher in tests with human evaluators than both Nano Banana 2 and the more expensive Nano Banana Pro. Independent tests have yet to be conducted. arynews
ElevenLabs v4: Voices with More Emotion
ElevenLabs creates AI voices: text in, spoken audio out. On 28 September, version 4 was released. The new model retains a voice better over long texts, speaks over 90 languages, and can clone a voice with just 10 seconds of audio. vktr
There is also a fast variant, v4 Turbo, for conversations with AI assistants. It responds in about 100 milliseconds. That’s fast enough to make a conversation feel natural. According to MindStudio, you can also try v4 for free. propakistanimindstudio
Runway: Seedance 2.5
Runway is a video platform that also offers models from others. Seedance 2.5 comes from ByteDance (the company behind TikTok). In Runway, it creates videos up to 30 seconds from a prompt plus up to 50 references, and it can edit or extend existing video. Thirty seconds at once is long for AI video. Most models stop sooner. Within Runway, the resolution only goes up to 720p. Source: Runway Help. runwaymlrunwayml
Claude: Chat and Cowork Become One
Cowork is the version of Claude that performs longer tasks independently: searching files, creating reports, organising things. Previously, this was separate from regular chatting. On web and desktop, Chat and Cowork now share a single start screen. Tasks run in the cloud, so they continue if your laptop is closed, and you receive a notification on your phone if Claude needs your approval. Source: The Decoder, Android Authority. the-decoderandroidauthority
Small
All 17,500+ voices from the ElevenLabs library work with v4, but older cloned voices need to be retrained. testingcatalog
To Follow
Independent tests of Nano Banana 2.1. So far, all figures are from Google itself.
According to one Taiwanese source, Nano Banana 2 will stop on 29 October. Not yet confirmed by Google. technews
