10 Prompts That Actually Fix Real Photos 📸

Welcome to another edition of Horizon AI,

In today's issue, we're sharing 10 AI prompts designed to help you fix and enhance real photos, from restoring old images and improving lighting to removing imperfections, upscaling quality, and making precise edits while preserving the original subject.

Let’s jump into it!

Read Time: 4.5 min

Here's what's new today in the Horizon AI

• China's Moonshot AI Releases the Largest Open-Source Model Ever, Rivaling Top U.S. Systems

• Canva Launches Code 2.0 for AI Website and App Creation

• AI Tutorial: Try These 10 Prompts to Fix Real Photos

• AI Tools to check out

• AI Findings/Resources

• The Latest in AI and Tech 💡

AI News

MOONSHOT

China's Moonshot AI Releases the Largest Open-Source Model Ever, Rivaling Top U.S. Systems

Chinese AI startup Moonshot AI stunned developers on Thursday with a massive new model that closes the gap with leading U.S. offerings and even surpasses OpenAI's and Anthropic's most capable systems on some benchmarks, all at a fraction of the cost.

Details:

• Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with a 1 million-token context window, multimodal reasoning capabilities, and an always-on reasoning mode for complex tasks.

• The model is designed to support a wide range of enterprise use cases, including coding, research, long-context analysis, and agentic workflows. It is also compatible with the OpenAI SDK, making it easier for developers to integrate into existing applications.

• In Kimi's own benchmarks, K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol on overall performance, though it surpasses them on some specific benchmarks. It consistently outperformed all other tested models.

• Notably, the model took the No. 1 spot in the Frontend Code Arena, surpassing Claude Fable 5; it also outranked the standard version of Anthropic's Opus 4.8 and tied Sol in the Arena's broader text ranking.

• Moonshot AI plans to release the model's open weights by July 27.

Moonshot is pricing K3 well below the premium models it's challenging (it costs $3 per million input tokens and $15 for output, the same as Anthropic's Sonnet 5), putting pressure on U.S. labs, which may soon struggle to justify charging top dollar for frontier-level intelligence.

Source

TOGETHER WITH RANGE

You're Invited: Tax-Smart Investing With Range Financial Planners

You've worked hard to fund your portfolio — your investment strategy should work just as hard to maximize your after-tax returns. On July 23, join Range's CFPs and CPAs live for the practical moves that put more of your investment returns back in your pocket.

What we’ll cover:

• Investment moves to maximize your after-tax returns

• How tax-loss harvesting can lower taxes due

• When direct indexing works (and when it doesn't)

• How to build a diversified portfolio that reduces tax drag

Range is all-in-one AI wealth management — tax, investments, retirement, and estate in one place. Bring your questions for the live Q&A. Free to attend, and seats are limited.

Reserve your spot today

^{This webinar is for informational purposes only and does not constitute investment advice or a recommendation to buy, hold, or sell any security. Forward-looking statements involve risks and uncertainties. Past performance is not indicative of future results. Range defines "high earners" as households with income over $300k.}

CANVA

Canva Launches Code 2.0 for AI Website and App Creation

Canva has introduced Canva Code 2.0, a major update to its AI coding tool that lets users create interactive websites, apps, and other experiences using natural language prompts.

Details:

• Canva Code 2.0 allows users to generate interactive websites, apps, calculators, forms, and other web experiences from text prompts, then edit them directly inside Canva using familiar drag-and-drop tools.

• The update introduces HTML import support, enabling users to bring projects created with other AI coding tools into Canva for further editing and customization.

• It also supports real-time collaboration between team members.

• Canva says the new version generates code 75% faster and reduces the time needed to publish a website by 30% compared to the previous release.

• More than 50 new interactive templates have been added, along with responsive layouts, mobile previews, and AI-powered editing to simplify website creation for non-technical users.

The feature is now available to all Canva users, including those on the free plan.

Source

AI Tutorial

Try These 10 Prompts to Fix Real Photos

Go to ChatGPT, select "Create an image," and try these:

1. Expression Change

Change the facial expression of the person in this photo to SMILING. Keep the person’s identity, age, bone structure, face shape, eye shape, nose, lips, and jawline exactly as in the original. Preserve skin texture, pores, freckles, moles, facial hair, and hairstyle. Adjust only the muscles involved in the new expression — mouth, cheeks, eyes, and eyebrows — so the change looks anatomically natural. If the expression involves a smile or laugh, show teeth only if appropriate and keep them consistent with the person’s real teeth shape and color. Preserve the original lighting direction, shadows, skin tone, and color grading. Do not change the head position, pose, gaze direction, clothing, or background. The result must look like a real photo of the same person captured a moment later with a different expression.

2. Natural Skin Retouch

Professional skin retouching with advanced imperfection correction. Remove pimples, acne, blackheads, temporary spots, redness, and under-eye darkness on facial skin. Intelligent smoothing that preserves natural pores, fine lines, and authentic skin texture, with no 'plastic skin' or airbrushed effect. Fully preserve the original skin tone, undertone, freckles, moles, and permanent beauty marks. Avoid excessive brightness, darkening, or whitening of the skin. Keep all facial details precise and untouched: eyes, eyebrows, eyelashes, beard, hairline, and individual hair strands. High-end beauty editorial finish with balanced lighting, refined contrast, and subtle facial definition. Studio-quality photorealistic result, ultra-detailed, natural skin texture, clean cinematic finish. Negative constraints: no face slimming, no eye enlargement, no lip reshaping, no skin smoothing beyond blemish removal, no beauty filter effects, no identity shift.

AI Tools to check out

🗣 Bono: Talk to Bono for 10 minutes. Get blog posts, social content, newsletters, and more — written in your voice.

🎬 HeyGen: Turn your ideas into videos in minutes. No cameras, no crew, no editing skills required.

💥 Ezier: Create images, videos, audio, and campaign-ready assets from one simple AI workspace.

🦾 Agata: Fix grammar, refine style, and translate text in any app on Windows and Mac.

🌐 SnapBlock: Create websites and apps in minutes with AI

AI Findings/Resources

🧠 Yuji Tachikawa, one of the world’s leading theoretical physicists, reports Claude Fable solved a problem that he and his collaborators had gotten stuck on for the past 6 months

🤯 Kimi K3 vs Claude Opus 4.8, same prompt

📌 5 tips and tricks for using NotebookLM

💡 Model and effort in Claude Code: knowing more vs. trying harder

⏳ ChatGPT reimagines what modern brands would look like in 1970s ads

The latest in AI and Tech

29 countries launch China-backed global AI cooperation organization

Twenty-nine countries have signed an agreement to establish the World Artificial Intelligence Cooperation Organization (WAICO), a China-backed international body focused on AI cooperation and governance. The organization aims to promote the safe, fair, and responsible development of AI while helping developing nations bridge the digital divide. The founding members are primarily from Asia, Africa, Latin America, and Eastern Europe; major Western countries are not part of the initiative.

Microsoft uses AI to deliver record 570 security fixes

The company has released 570 security patches across Windows, Office, and other products, marking its largest monthly security update to date. It said the increase is partly due to its use of AI to identify software vulnerabilities that might otherwise have gone undetected.

Google's NotebookLM is now Gemini Notebook

Google has renamed NotebookLM to Gemini Notebook, aligning its AI research tool with the broader Gemini brand. Alongside the rebrand, Google introduced new capabilities, including native code execution for data analysis, and announced upcoming integration with AI Mode in Google Search.

Google faces another lawsuit alleging its AI violated copyrights

The lawsuit was filed in New York by Hachette Book Group, Cengage Learning, Elsevier, author Scott Turow, and his publishing company S.C.R.I.B.E.

They accuse Google of using copyrighted books to train its Gemini AI models without permission, and say the company's actions are harming authors and the wider publishing industry, negatively impacting book sales.

Gemini in Chrome rolling out to the UK

The company is expanding Gemini in Chrome to users in the United Kingdom, more than a year after its initial launch. The rollout does not yet include users in the European Union.

That’s a wrap!

Thanks for sticking with us to the end! Let’s stay connected on LinkedIn and Twitter.

Not subscribed yet? Sign up here and send it to a colleague or friend!

See you in our next edition!

Gina 👩🏻‍💻