- NATURAL 20
- Posts
- ChatGPT Adds Image Editing
ChatGPT Adds Image Editing
PLUS: Meta Glasses Get Hearing Updates, Adobe Adds Video Editing by Text and more.
Wes Roth
December 17, 2025
In partnership with
The Future of Shopping? AI + Actual Humans.
AI has changed how consumers shop by speeding up research. But one thing hasn’t changed: shoppers still trust people more than AI.
Levanta’s new Affiliate 3.0 Consumer Report reveals a major shift in how shoppers blend AI tools with human influence. Consumers use AI to explore options, but when it comes time to buy, they still turn to creators, communities, and real experiences to validate their decisions.
The data shows:
Only 10% of shoppers buy through AI-recommended links
87% discover products through creators, blogs, or communities they trust
Human sources like reviews and creators rank higher in trust than AI recommendations
The most effective brands are combining AI discovery with authentic human influence to drive measurable conversions.
Affiliate marketing isn’t being replaced by AI, it’s being amplified by it.
Hey there!
There are days where it feels like “AI news” is mostly vibes.
And then there are days like this, where the updates are very practical.
Today:
OpenAI shipped a new version of ChatGPT Images on December 16, 2025, and it’s basically a “make changes without breaking the whole photo” leap.
What stood out to me:
Edits are more surgical. You can ask for a specific change and it tries hard to keep lighting, composition, and people’s appearance consistent across edits.
It’s up to 4× faster, which matters more than people think—iteration speed is the whole game for creative work.
Text rendering is meaningfully better, including denser/smaller text (think menus, posters, “infographic-y” images).
There’s a new Images space in the ChatGPT sidebar with preset filters + trending prompts, and even a one-time likeness upload so you can reuse your look without digging through your camera roll.
If you use the API: it’s available as GPT-Image-1.5, and OpenAI says image inputs/outputs are 20% cheaper than the prior version.
Google Labs introduced CC, an experimental productivity agent built with Gemini.
The pitch is simple (and kind of compelling if you live in your inbox):
It can connect to Gmail, Google Calendar, Google Drive, and the wider web to understand your day.
It sends a “Your Day Ahead” briefing every morning that summarizes schedule + tasks + updates.
It can also prep email drafts and calendar links, and you can steer it by replying with custom requests (including “remember this”).
Xiaomi posted MiMo-V2-Flash on Hugging Face under an MIT license.
Why it’s notable (even if you don’t plan to run a 309B model at home):
It’s a Mixture-of-Experts model with 309B total params / 15B active—classic “big brain, smaller per-token compute” strategy.
It targets high-speed reasoning and agentic workflows, using a hybrid attention setup (sliding window + global attention) to reduce KV-cache storage, plus Multi-Token Prediction to boost decoding speed.
It claims up to 256k context and mentions training on 27T tokens with FP8 mixed precision.
They also say they’re open-sourcing 3-layer MTP weights to encourage community research.
This one feels like it’s aimed at the “agents + long context + throughput” crowd more than mainstream chat use.
🧠RESEARCH
ReFusion accelerates text generation by planning chunks of sentences simultaneously rather than writing one word at a time. This hybrid approach combines the speed of broad planning with the precision of detailed writing, delivering 18 times faster speeds and higher quality than previous models while using computer memory more efficiently.
Standard tools that translate images into AI code focus too much on pixel-perfect copying, which limits creativity. VTP fixes this by teaching the system to understand the actual meaning and context of an image. This shift drastically boosts generation quality and efficiency, proving that "understanding" matters more than just "mimicking."
QwenLong-L1.5 is designed to handle massive documents containing millions of words. Instead of trying to process everything at once, it uses a smart "memory agent" that reads and takes notes iteratively. This method allows it to solve complex reasoning puzzles in long texts, rivaling the performance of top-tier models like GPT-5.
🛠️TOP TOOLS
Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.
AssemblyAI : Speech AI API Platform - Speech AI platform that turns audio/video into accurate text
Assets Scout – AI-Powered 3D Asset Search Tool - AI‑powered visual search engine for 3D content.
Astria AI – Generative Imaging and Photography - a platform for creating personalized visuals by fine‑tuning leading diffusion models on a handful of your reference photos, then generating consistent brand styles.
📲SOCIAL MEDIA
🗞️MORE NEWS
Meta Smart Glasses Updates Meta updated its smart glasses to help you hear conversations better in noisy places by amplifying the voices of people you are looking at. The update also lets you stare at an object, like an album cover, and ask the glasses to play related music on Spotify. These features are currently rolling out to early testers in the US and Canada before a wider release.
Adobe Firefly Video Editing Adobe’s new video tools let you edit specific parts of a video, like changing colors or objects, just by typing a description of what you want. They also added support for AI models from other companies, giving creators more options for generating images and improving video quality within Adobe’s software. This update moves video editing closer to being as easy and flexible as writing a text message.
Meta SAM Audio Meta released a new AI tool called SAM Audio that can pick out and separate specific sounds from a recording. You can remove unwanted noises or isolate a voice simply by typing what you want to hear or clicking on the sound’s source in a video. This makes fixing bad audio or creating cleaner sound mixes much faster and easier for creators.
OpenAI Frontier Science OpenAI launched a new program to turn AI into a helpful partner for scientists and mathematicians. Their latest model, GPT-5.2, has already helped researchers solve complex math problems and improve biological experiments, such as finding better ways to clone DNA. The goal is to speed up scientific breakthroughs by letting AI handle some of the heavy mental lifting and analysis.
Oracle’s $248 Billion Bet Oracle has committed to paying a massive $248 billion in future rent for data centers, mostly to support OpenAI’s computing needs. Financial experts worry this is a risky bet, as Oracle owes this money even if the current AI boom slows down or the startups renting the space fail. This massive debt could become a serious problem if the demand for AI computing doesn't last as long as the leases.
OpenAI Executive Departure Hannah Wong, the head of communications at OpenAI, is leaving the company after five years. She played a key role in managing the company's public image during the launch of ChatGPT and the chaotic firing and rehiring of CEO Sam Altman. Her departure marks another major leadership change as the company continues to grow and face intense public scrutiny.
AI Heart Disease Detection Doctors have developed an AI tool that can spot a hidden form of heart disease using a standard 10-second heart rhythm test (EKG). Usually, this condition requires expensive and invasive scans, but the AI can detect it just by analyzing the electrical patterns of the heartbeat. This innovation could help millions of patients get a correct diagnosis quickly, cheaply, and without special equipment.
What'd you think of today's edition? |
Reply