Firefly Video vs. Gen-3, Luma & MiniMax
Which AI Video Tool Delivers the Best Results?
Total reading time around 5 minutes.
Welcome to Visually AI!
🔮AI News this Week
Adobe Firefly Video
Adobe’s upcoming Firefly Video Model offers powerful tools for film editors and video professionals, integrating seamlessly into Adobe Creative Cloud, Premiere Pro, and Adobe Express later this year.
Key features include:
Text to Video & Image to Video: Allows users to create videos from text prompts or animate still images.
Generative Extend: Designed to fill gaps in timelines, extend video frames, and add new elements to existing footage.
The Generative Extend feature is especially useful for editors needing to hold shots longer or create smooth transitions. The ability to extend and seamlessly edit videos enhances storytelling potential while maintaining creative freedom.
Here is an example of Generative Extend, from Adobe:
How does Adobe Firefly Video compare to other AI video models?
It’s hard to believe how many AI video models are currently available, so I decided to test the prompts Adobe provided in the announcement. I ran the same prompts in Runway’s Gen-3, Luma AI Dream Machine, and Hailuo AI’s MiniMax.
The results were surprising! Here are a few examples:
Prompt: Cinematic closeup and detailed portrait of a reindeer in a snowy forest at sunset. The lighting is cinematic and gorgeous and soft and sun-kissed, with golden backlight and dreamy bokeh and lens flares. The color grade is cinematic and magical.
Prompt: Macro detailed shot of water splashing and freezing to spell the word "ICE”.
Prompt: Hand-drawn simple line art, a young kid looking up into space with a wondrous expression on his face.
We don’t know when Firefly Video will be available, yet. If you have an existing Adobe account, you can sign up for the waitlist here.
OpenAI’s o1 Models
OpenAI launched a new series of AI models called “o1-preview” and “o1-mini,” with improved reasoning capabilities for solving complex problems in math, coding, science, and more.
The models are slightly different, according to OpenAI’s announcement email:
The o1 models are currently available in ChatGPT and OpenAI’s API, for ChatGPT Plus and Team users. Enterprise and Edu users will get access to both models next week, and they stated that o1-mini will eventually be open to all Free users.
You’ll need to select these models manually and there are weekly rate limits capped at 30 messages for o1-preview and 50 for o1-mini:
Google’s NotebookLM Audio Overview
Google Labs gave us a new way to quickly understand complex information with Audio Overview. Upload source material, such as research papers, presentation slides, or any documents and turn it into a podcast-style discussion between two AI hosts.
Go to NotebookLM
Create a “New Notebook”
Upload a document or source information
Click “Generate” in your Notebook guide
I uploaded a new research paper titled MEDIC:Towards a Comprehensive Framework for Evaluating LLMs in Clinical Applications by Kanithi et al., 2024. It seemed perfect for this test!
It takes about 3-5 minutes and the episodes are quite engaging. I listened to the entire conversation and understand the researchers’ points, with the ‘AI hosts’ even thanking the listeners at the end!
Listen to the full conversation here, in an audiogram I created in Canva and posted on 𝕏 and LinkedIn.
Here’s a quick demo:
📸AI Snapshots
OLMoE, a new open-source large language model, is making waves with its impressive performance and efficiency. Developed by AI2 and Contextual AI, OLMoE uses a "Mixture of Experts" (MoE) architecture, where a team of specialists work together to complete tasks. A "gating network" manager selects the best experts for each task, combining their outputs for a final result.
OLMoE stands out as fully open-source and achieves comparable performance to larger models while using fewer parameters (1 billion out of 7 billion). This makes it a game-changer for deploying AI on less powerful hardware, making AI more accessible to a wider range of users.
Y Combinator is doubling its annual cohorts from two to four starting in 2025, aiming to provide more frequent and personalized support to founders, including those in AI, by reducing cohort sizes while maintaining the overall number of participating startups at approximately 500 per year. (Bloomberg - paywalled)(Perplexity Daily)
Magnific added a new Grid view to show your previous upscales. You can also choose favorites by clicking on the heart in the top of each expanded image, then view all favorites in a Grid view alone. (link)
🛠️ This Week’s AI Tools
VideoGen: AI-powered video generator that allows users to create professional, copyright-free videos with realistic AI voice-overs from text prompts, featuring 150 unique voices across 50+ languages and accents. (link)
Best Replit Agent AI Apps: Free directory of apps built by Replit Agents organized by category and updated daily. (link)
Digital Carbon: AI transforms images into 3D immersive experiences for e-commerce product showcases, virtual tourism, and more. (link)
Anifusion: Generate comics and manga from text prompts using the canvas editor to create full stories. (link)
GPT Engineer: Chat with AI to build web apps 10x faster, sync with GitHub and deploy with one-click. Builds front-end with React, Tailwind & Vite. (link)
Latent Navigation: Hugging Face Space by Linoy Tsaban and apolinario lets you explore CLIP text space between 2 opposite styles or concepts with FLUX.1 schnell. Enter a simple prompt and 2 directions to steer style or concepts, then click Generate. (link)
Example:
Prompt: hyper-realistic pixelated fish
Directions to steer: pixar -> ukiyoe
Watch the transition from pixar to ukiyoe:
🎧IntelliVerse Podcast & Visually AI on YouTube
I had a fascinating conversation with Adam B. Levine, Co-founder & CIO of Blockade Labs on the latest IntelliVerse podcast episode.
It’s available on YouTube, Spotify, Apple Podcasts, and PodBean.
You can watch the full interview here:
🖼️ Image Prompts
Prompt: A luxurious container of artisanal organic body butter on a rough-hewn marble surface, surrounded by delicate, dew-kissed rose petals. A crumpled antique linen cloth drapes nearby. Soft, golden hour light filters through sheer curtains, casting a warm glow. Hyper-realistic style with intricate textures and reflections.
I generated this image in Midjourney, using style reference code (sref) 698401885:
Prompt: Simple white cotton blouse, front view, crisp details, minimalist fashion, soft studio lighting, neutral background
Thanks for reading and have a creative week!