Welcome back!

The August surge of text-to-image models continues with the launch of Ideogram 2.0, claimed to outperform DALL·E 3 and Flux.1 in benchmarks. Its advanced text generation capabilities make it a game-changer for infographics and marketing materials. Let's unpack...

Today’s Summary:

• 🎨 Ideogram 2.0 frontier image model

• 💻 Microsoft Phi Mini tops benchmarks

• 🛠️ Google adds free Prompt Gallery to AI Studio

• ✨ Adobe Magic Fixup enhances photo editing

• 🎓 Claude supports LaTeX math formulas

• 🌐 Midjourney web offers 25 free images

• 💬 Elevenlabs cuts prices by 50%

• 🛠️ 2 new tools

TOP STORY

Ideogram 2.0 Image Generator beats Dall-E 3 and Flux Pro in benchmarks

The Summary: Ideogram has unveiled Ideogram 2.0, a major update that claims to surpass DALL·E 3 and Flux Pro in human evaluations. The new model is particularly strong in text generation and offers a public API beta along with an iOS app.

Key details:

• Ideogram 2.0 rated higher image quality than DALL·E 3 and Flux Pro in human evaluations

• 40 free images per day (public generations)

• New text-to-image API (beta)

• Upgraded model includes new styles, color palette control, tools for creators

Why it matters: As competition in image generation heats up, Ideogram 2.0 stands out for its superior text generation capabilities. This makes it especially well-suited for design tasks like marketing materials, book covers, posters, and product packaging.

Try it here: Web platform - iOS app

MICROSOFT

Microsoft Phi 3.5 Mini models deliver exceptional performance

The Summary: Microsoft has released three impressive Phi open-source AI models that deliver outstanding performance despite their compact size. These models achieve top scores across various benchmarks for tasks such as code generation, mathematical reasoning, and multimodal understanding.

Key details:

• The compact 3.8B parameter mini-instruct beats LLama-3.1-8B

• The 16 x 3.8B MoE-instruct (6.6B active parameters at inference) surpasses Gemini Flash

• The 4.1B parameter Phi-vision-instruct outperforms Claude Sonnet 3.5-vision and is comparable to GPT-4o-vision

• The models are available on Hugging Face

Why it matters: Microsoft's Phi-3.5 AI model series pushes the limits of efficiency for mini models, making them compelling open-source alternatives (MIT licensed) for a wide range of on-device tasks.

Try it here: Phi 3.5 mini - Phi 3.5 MoE - Phi 3.5 vision

TOOLS

Google Launches Free Prompt Gallery in AI Studio

The Summary: Google has added a free Prompt Gallery to its AI Studio platform, providing developers with a library of pre-built prompts for its Gemini models. This update includes tools for long context, multi-modal inputs, and structured outputs.

Key details:

• Covers applications like math tutoring, recipe generation, and trip planning

• Supports image, video and audio inputs for multi-modal AI tasks

• Includes technical tools such as code unit testing

• Available at no cost, lowering barriers to AI development and experimentation

Why it matters: Google’s continued offering of free advanced AI tools could pressure other providers to reconsider their pricing models to remain competitive.

Try it here: aistudio.google.com

QUICK NEWS

Quick news

• Adobe Magic Fixup can erase items from photos while also adjusting lighting and shadows

• Claude supports rendering LaTeX math formulas

• Midjourney web interface offers 25 free images

• Elevenlabs is now 50% cheaper (1 character = 0.5 credits)

TOOLS

🥇 New tools

• Zed - Next generation code editor with AI assistance

• Boggl - Create product requirements and release notes with AI