V1 transforms images through automatic animation or manual prompts, where users can describe specific camera movements and actions.

Each job creates four 5-second clips extendable to 20 seconds, priced at 8x image costs, which Midjourney says is 25x cheaper than rivals.

V1 can handle images from both Midjourney and external options, with video outputs having the signature feel found in the startup’s image models.

CEO David Holz said V1 is a stepping stone towards real-time open-world simulations, which require the building blocks of image, video, and 3D models.

It now gives “smarter” answers, follows instructions more reliably, and offers better performance in longer conversations.

For complex questions, ChatGPT now runs multiple searches in parallel before responding.

You can now upload images, and ChatGPT will search the web based on what's in those images, adding a multimodal search tool to the platform.

Internal testing shows people are finding the new search more useful than before, even though it sometimes gives answers that are longer than necessary.

The update is available to all users, but the company notes that it may still make occasional mistakes, so double-checking responses is still necessary.

SEAL allows models to generate their own "self-edits" — instructions for creating synthetic data and setting parameters to update their own weights.

It learns through trial-and-error via a reinforcement learning loop, rewarding the model for generating self-edits that lead to better performance.

In knowledge tasks, the AI learned more effectively from its own notes than from learning materials generated by the much larger GPT-4.1.

The system also dramatically improved at puzzle-solving tasks, jumping from 0% with standard methods to 72.5% after learning how to train itself effectively.

Two AI-generated hosts promoted 133 products in the session, showcasing items while utilizing human gestures and handling real-time viewer interactions.

The stream reached 13M viewers and beat Luo’s “real” stream in May in just 26 minutes, with Baidu’s ERNIE crafting 97K+ characters of product descriptions.

Baidu said the stream was the first to feature ‘dual digital avatars’, with Luo and his digital co-host interacting in natural conversation and movements.

Over 100k digital humans reportedly work in China's $946B live commerce sector, slashing costs by 80% and increasing transactions by 62% on average.

Go to Higgsfield and sign up with your account.

Click on ‘Create>Image’

Upload your avatar and product image.

Click on ‘Generate’ to make your custom AI avatar hold your product.

To convert it into a video, click on the three dots in the right corner and select ‘Speak’.

Select a motion preset to define your avatar's movement and your audio.

Once added, click on the ‘Generate’ button to get your product video.

Open the Gemini app and tap the Plus button in the Ask Gemini box.

Tap Files and select a document either from your device or Google Docs (DOC, DOCX, PDF, RTF, and TXT formats are supported).

Once the file has been uploaded, tap Generate Audio Overview. (It may take a few minutes, but you don’t have to stay in the chat — you’ll be notified when it’s ready.)

Once it’s done, tap Audio Overview to listen to it.

You can find the Audio Overview at any time by tapping the Chats and Gems icon in the top-left corner of the screen.

You can also share and download the audio. From the playback screen, tap the three-dot icon in the top-right corner, then tap Download or Share.

OpenAI rolled out updates to its “Projects” feature, with new support for deep research and voice mode alongside improved memory functionality.

TikTok released new updates to its ‘Symphony AI suite’, including image-to-video, text-to-video, and AI avatar marketing for advertising content.

ChatGPT’s “Canvas” tool now supports downloads, letting you export projects as PDFs, Word docs, markdowns, or a variety of coding formats.

Reddit launched a pair of AI tools that let advertisers track what’s trending in different subreddits and test-drive new campaigns before rolling them out on the platform.

ElevenLabs now supports ‘MCP’ (model context protocol), meaning you can now build voice agents that connect to Salesforce, HubSpot, and Gmail without any setup.

Perplexity launched new video generation capabilities, enabling users to generate Veo 3 videos with audio on social media by tagging the @AskPerplexity account.

Google just graduated its ‘Gemini 2.5 Pro and Flash’ models to stable production status after months of preview version updates, while also rolling out a new, hyper-efficient ‘Flash-Lite’ variant.

MiniMax debuted ‘Hailuo 02’, a new AI video model that moves to No. 2 on the Artificial Analysis leaderboard, passing Veo 3.

Adobe released new mobile apps for its Firefly platform, allowing users to access its AI image, video, and other creative tools via iOS and Android.

Higgsfield released ‘Canvas’, a new image editing model with advanced inpainting controls for adding products or quickly changing details of an output.

Google introduced ‘Search Live with AI Mode’, allowing users to chat with a Gemini-powered voice search, receive spoken answers, and see linked sources in real-time.

OpenAI rolled out ‘ChatGPT Record’, a new feature allowing the assistant to capture, summarize, and transcribe audio from meetings and brainstorms.