LLMs, Image & Video Generation Just Got Even Better
Discover the latest advancements in AI with even more powerful LLMs, cutting-edge video models, and the best tools for image generation
Reading time about 5 minutes.
Welcome to Visually AI!
OpenAI GPT-o3
In the last edition of Visually AI, I wrote about DeepSeek challenging ChatGPT’s GPT o1. This week, we have another LLM with incredible capabilities: Alibaba’s Qwen2.5-Max.
This new model showcases incredible reasoning, multi-turn dialogue capabilities, and improved instruction-following performance, proving to be a strong competitor in the evolving AI landscape. As more models emerge, the competition is pushing generative AI capabilities to new heights.
You can see a comparison below:
Adobe Firefly Video is here!
Finally, Adobe Firefly Video is here, and it is a powerful video model with text to video and image to video capabilities, including using first and last frames to create a full scene.
It can create portrait (9:16) or widescreen landscape (16:9) layouts with optional settings, such as Shot size, Camera angle, Motion, and using Seed numbers to recreate a similar video.
Here is an example of my text to video result, using this prompt in Firefly Video:
“[Tilt up, Left circling] From the base of an ancient pyramid as storm clouds gather, lightning illuminating hieroglyphics. [Zoom out] Sand swirls in the wind. Epic scale, dramatic lighting.”
But - Firefly Video pricing might be unaffordable for most people, and it requires a separate Firefly subscription:
How to write better AI video prompts: Nature scenes
Prompting for nature scenes seems complicated but you really only need to describe what you want to see in your video.
If you have an input image, think about what would actually occur in real life and include the details in your prompt:
Subject and surrounding areas
Time of day, lighting, and weather
Environmental interaction with objects from wind, rain, snow, etc.
Finally, describe the type of camera motion to complete the scene.
I used these prompts to animate my Midjourney images with Kling 1.6:
Cinematic wide shot of a beautiful contemporary lakefront cabin in a redwood forest on a sunny day. Camera pushes in very slowly as the sunlight filters through the tree canopy and reflects off the lake water. A light breeze blows the trees and water.
Luma Ray2 Image to Video
Ray2 by Luma AI was already a great text to video model, but the new image to video feature is mind-blowing:
Realistic physics
More natural motion
Improved coherence & quality
I’ve been testing the new model with early access and it is extremely impressive. It’s available now for all unlimited plans and other subscription plans will be added soon. Additional features are on the way, including Start & End frames, Extend video, and Loops.
Here’s a quick look at my first results using a Midjourney image:
📸 AI Snapshots
Mistral unveiled the latest updates to its multi-modal le Chat, with code interpreter, extremely fast reasoning and analysis, and image generation powered by Black Forest Lab’s FLUX model. Available online, iOS, and Android apps.
Ideogram has a new Text Tool in Canvas that lets you add text and choose multiple fonts and colors. It’s available for all users.
Here’s a demo from Ideogram:
Pika introduced Turbo Mode, which generates 3x faster, uses less credits, with the same quality. Pika also has a new way to add anything or anyone to any video with Pikadditions:
Krea Chat is available in open beta right now, and it lets you generate and edit images by chatting with the model.
💻 Recently On Visually AI Youtube
LumaLabs Ray2 Showcase | What can it produce? (No Editing)
📱 Recently On 𝕏
Prompting for nature scenes seems complicated but you really only need to describe what you want to see in your video. If you have an input image, think about what would actually occur in real life and include the details in your prompt:
• Subject and surrounding areas
• Time of day, lighting, and weather
• Environmental interaction with objects from wind, rain, snow, etc. Finally, describe the type of camera motion to complete the scene.
I used this prompt to animate my Midjourney image with Kling 1.6: Cinematic wide shot of a beautiful contemporary lakefront cabin in a redwood forest on a sunny day. Camera pushes in very slowly as the sunlight filters through the tree canopy and reflects off the lake water. A light breeze blows the trees and water.
-
Midjourney -> Luma Ray2 -> MMAudio
-
Don't be afraid to step outside of your comfort zone.
You never know what you might achieve.
-
Midjourney -> Kling 1.6 -> MMAudio
🌎 AI Developments All Over The Globe
1. Elon Musk Attempts to buy OpenAI with $97.4 billion bid
Elon Musk has led a consortium in proposing a $97.4 billion bid to acquire OpenAI’s nonprofit arm. Musk, who co-founded OpenAI in 2015 but departed in 2018, aims to return the organization to its original nonprofit mission, expressing concerns over its shift towards a for-profit model.
OpenAI’s CEO, Sam Altman, has dismissed the offer, stating that the company is not for sale and suggesting that Musk’s bid is a tactic to disrupt OpenAI’s operations. Altman emphasized that OpenAI’s structure is designed to prevent any single individual from taking control.
2. EU Launches InvestAI with €200 Billion for AI Development
The European Commission has announced a €200 billion investment plan named InvestAI, aimed at advancing artificial intelligence across Europe. This initiative includes €20 billion allocated for “AI gigafactories” to assist companies in training complex AI models.
3. MIT Chemists Use Generative AI to Predict 3D Genomic Structures
Researchers at MIT have developed a generative AI system capable of predicting how specific DNA sequences arrange themselves within the cell nucleus. This approach accelerates the process, taking minutes rather than days, and holds promise for advancements in genomics.
4. Investors Recognize Alibaba as Emerging AI Leader
Alibaba has seen a significant increase in its enterprise value, rising by nearly $87 billion, as investors view the company as a new leader in the AI sector. This surge reflects growing confidence in Alibaba’s AI capabilities and potential.
5. Advancements in AI for Medical Diagnostics
Recent developments have demonstrated the use of machine learning to generate synthetic medical images, such as x-rays, to augment AI training. This approach aims to improve diagnostic accuracy by providing AI systems with a more diverse set of training data.
🚀 My Recent Top AI Tools Picks
Replit iOS app: Build full apps and deploy in minutes from your iPhone.
Nim Video: Full-service AI video platform with many of the top models, including Kling Pro, Hunyuan, Minimax, and several features like character reference, video relighting and more.
Hugging Face Spaces: Quickly search for the app you need in Hugging Face Spaces with over 400k available.
🖼️ Image Prompts
Prompt: A tall, black, vertical house, placed in the middle of a forest with a lake beside it. Hyper-realistic rendering, cinematic lighting, natural light, architectural photography, highly detailed.
Prompt: The silhouette of an ancient Roman temple at sunset, with distant buildings visible through an open door in front of it. The sun is setting behind the building, casting long shadows across its surface.
🎥 Video Prompt
Prompt: A crystalline deer steps through a forest where the trees are made of flowing liquid metal. Sunlight filters through the chrome canopy. Camera gently pushes in, ultra smooth coherent movement.
From Pika 2.1 Turbo
Thank you for reading.
Have a creative week!
Great article as always, Heather. The potential for creators to visualize concepts in real-time is huge, but yeah, that pricing might make it a little out of reach for many. Here's hoping Adobe offers more flexible options in the future.