• Innov8 AI
  • Posts
  • ChatGPT- ഡെ new image model വരുന്നു ..

ChatGPT- ഡെ new image model വരുന്നു ..

PLUS: NotebookLM new features | L8R by Innov8

Author

Innov8
December 10, 2025

In partnership with

Rumours say OpenAI is secretly roasting Chestnuts and Hazelnuts. 🌰

But here’s the real tea ☕:
OpenAI seems to have leaked a new image model that does something old models always failed at

clean code inside images.

No more alien letters when you ask for Python snippets in a PNG.

At the same time, Microsoft quietly dropped an AI that turns a ₹800 tissue slide into a cancer-fighting data machine.

Google’s NotebookLM is slowly turning into that toppper friend who speaks 80 languages and helps you finish homework.

So grab your snacks

I’m Alex. Welcome to L8R by Innov8.
Let’s DIVE DEEP 🧠👇

In today’s post:

  • OpenAI’s new image model can now draw actual, readable Python code inside an image

  • Microsoft’s new AI can convert a ₹800 slide into a cancer research powerhouse

  • NotebookLM now talks in 80 languages and behaves like a teacher who never gets tired

GPT-Image-2 (Chestnut & Hazelnut)

People in AI forums are talking about two secret OpenAI image models: “Chestnut” and “Hazelnut.”

They say these models can:

  • Make celebrity-style selfies

  • Put real code and formulas inside images

  • Mix anime + pencil sketch + claymation in one single picture

Nothing is official yet. But the leaks are already making artists nervous and excited at the same time.

📰 What’s New?

Two secret models in tests

  • Testers on sites like Design Arena and LM Arena found two new OpenAI image models.

  • They call them Chestnut (small / mini model) and Hazelnut (bigger main model).

  • People think Chestnut is like “Image-2-mini” and Hazelnut is the full “Image-2” model.

Better world knowledge & selfies

  • Early scores say these models know real-world facts almost as well as Google’s top image model.

  • They can make super realistic selfies with good lighting, skin tone, and face shape.

  • Quality looks better than GPT-Image-1.

Real code and text inside images

  • Old image models often made broken text or gibberish code.

  • Chestnut/Hazelnut can draw readable code, flowchart labels, and math formulas inside images.

  • This is useful for slides, diagrams, and UI mockups.

Cleaner structure and colours

  • Objects in the image stay in the right shape and place.

  • Colours look natural and balanced, not weird or over-saturated.

  • Style is more consistent across a whole image.

Built on GPT-4o base

  • Testers think these image models are built on GPT-4o tech.

  • But they still feel like a big jump over GPT-Image-1 (released in 2025).

🚨 Why It Matters

  • Better tools for normal people
    Anyone can make pictures that look like pro photos, comics or posters, just by typing a prompt.

  • Huge win for devs and teachers
    Clear code blocks, charts, and math inside images help with teaching, docs, and app design.

  • AI image wars are heating up
    Google’s Nano Banana Pro and ByteDance’s USO are already strong.
    These leaks show OpenAI is trying to climb back to #1.

  • Hint of future GPT
    People think Chestnut/Hazelnut may ship with GPT-5.2, showing a tighter link between text and images.


From our Partners

From Boring to Brilliant: Training Videos Made Simple

Say goodbye to dense, static documents. And say hello to captivating how-to videos for your team using Guidde.

1️⃣ Create in Minutes: Simplify complex tasks into step-by-step guides using AI.
2️⃣ Real-Time Updates: Keep training content fresh and accurate with instant revisions.
3️⃣ Global Accessibility: Share guides in any language effortlessly.

Make training more impactful and inclusive today.

The best part? The browser extension is 100% free.

Microsoft GigaTIME — Cancer Research AI

Microsoft’s new AI model GigaTIME can look at a cheap tissue slide (around $10) and guess super-detailed immune system data that usually needs expensive lab tests.

It’s like turning an old black-and-white TV into a 4K smart TV using only software.

📰 What’s New?

What is GigaTIME?

  • Full name: Gigapixel Tumor MicroEnvironment.

  • It looks at normal H&E slides (the standard pink-and-purple slides pathologists use).

  • It predicts what a fancy lab test called multiplex immunofluorescence (mIF) would show.

  • It can predict 21 different protein channels for each slide.

Huge training dataset

  • Trained with Providence Health in the US.

  • Used 40 million cells where each cell had:

    • A normal H&E image

    • A matching real mIF image.

Large-scale testing

  • Then applied to 14,256 cancer patients.

  • Data came from 51 hospitals and 1,000+ clinics.

  • It generated about 300,000 virtual mIF slides.

  • Covered 24 cancer types and 306 cancer subtypes.

New patterns found

  • GigaTIME helped find 1,234 important links between:

    • Protein activity

    • Cancer type, stage, or patient survival.

  • For example, links between:

    • Proteins like CD138, CD20, CD4

    • Genes like KRAS and KMT2D

    • And how many immune cells are inside a tumor.

External check

  • They tested the model on a separate group of 10,200 patients from TCGA (a big cancer database).

  • The predicted protein values matched real lab results with a high correlation (0.88).

Open-source

  • Microsoft released GigaTIME for researchers on Foundry and Hugging Face.

  • This means many labs can build on top of it.

🚨 Why It Matters

  • Big cost drop

    • Real mIF lab test: about $550 per slide.

    • Normal H&E slide: about $5–10.

    • GigaTIME can turn cheap slides into virtual rich data in ~20 minutes per slide on a single GPU.

  • Faster research

    • Old style: only small groups of patients can be tested.

    • New style: hundreds of thousands of slides can be re-used as rich data.

    • This speeds up new biomarker discovery.

  • Better understanding of immunity

    • Helps scientists see where immune cells live inside tumors.

    • Helps guess who may respond better to treatments like immunotherapy.

  • Step toward “virtual patients”

    • GigaTIME’s data could combine with genomics, scans, and notes.

    • This can help create digital twins of patients to simulate treatments.

Together with guidde

Find customers on Roku this holiday season

Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

NotebookLM gets NEW updates

Remember that one smart kid in class who:

  • Takes great notes

  • Explains things clearly

  • Helps everyone before exams?

Now Google is trying to make that kid into an AI app called NotebookLM.
And it keeps getting smarter — it now finds sources, reads PDFs with images, makes flashcards, quizzes you, talks in 80+ languages, and even acts like a personal tutor.

📰 What’s New?

1. Discover Sources & Multimodal PDFs

  • You can click “Discover sources” and NotebookLM will find related web pages and add them to your notebook.

  • It can read PDFs with text, images, and graphs together.

  • Available in many countries and supports dozens of languages.

2. Video & Audio Overviews in 80 Languages

  • NotebookLM can create short video summaries of your notes.

  • These video and audio overviews now support around 80 languages.

  • Non-English audio summaries are now as good and detailed as English ones.

3. Flashcards, Quizzes & Learning Guide

  • It can turn your notes or textbook pages into:

    • Flashcards

    • Multiple choice questions

    • Practice quizzes

  • Each card has an “Explain” button with more detail and citations.

  • A new mode called “Learning Guide” acts like a tutor:

    • Asks you questions

    • Breaks topics into steps

    • Makes you think before showing the answer.

4. More report formats & fun audio modes

  • Reports can now come out as:

    • Blog posts

    • Research proposals

    • Whitepapers

  • Audio Overviews got new modes like “Critique” and “Debate” where two AI voices discuss a topic.

5. Deep Research agent & more file types

  • A new Deep Research mode can handle big, complex questions.

  • It reads your files plus online sources and builds a structured report with citations.

  • You can upload spreadsheets, slide decks, and other file formats, not just PDFs or docs.

6. Custom Personas & Bigger Instructions

  • You can now give up to 10,000 characters of custom instructions.

  • This lets you set NotebookLM as:

    • A product manager

    • A teacher

    • A research assistant

    • Or any custom role you want

  • It also remembers more of your past chat for better context.

🚨 Why It Matters

  • Better study partner

    • It doesn’t just summarize. It teaches, quizzes, and guides you.

  • Works for many languages

    • 80+ languages means students around the world can use it, not only English speakers.

  • Helps with real research

    • Deep Research mode + Discover Sources = faster literature review style work.

  • More personal

    • Custom personas and big instruction fields mean NotebookLM can match your style and goals.

  • Active learning

    • Flashcards, quizzes, and Learning Guide make you think and respond, which helps you remember more.

🚨 Quick L8R Summary

  • OpenAI's Secret Snacks: Leaked tests reveal two new OpenAI image models (codenamed "Chestnut" and "Hazelnut") that can finally write perfect code and math formulas inside images.

  • Microsoft's $10 Cancer Breakthrough: A new AI called GigaTIME turns cheap, standard microscope slides into rich, expensive lab data, speeding up cancer research massively.

  • The Ultimate AI Tutor: Google's NotebookLM just got a huge update—it now speaks 80+ languages, summarizes videos, and creates quizzes to help you study.

Ithrollu innathe AI Update.

appo adutha l8ril varam.. bie.

Innathe L8R engane ondarunnu?

Login or Subscribe to participate in polls.

Founders, Coaches & Business Owners - Listen up!

You're the Bottleneck in Your Content System

Most founders reading this know content matters. But between calls, product work, and actually running the business where's the time to record, edit, and post?

Here's the truth: your personal brand can grow without you being in front of a camera daily.

I run the Founder's Clone Challenge an 8-day system that teaches you to build an AI-powered content workflow. By Day 8, you have a digital clone (your voice, your face, your style) that your editor can use to create daily content while you focus on what actually moves your business forward.

100+ founders have already built theirs. Batch 6 starts on December 16.