Top 100 Gen AI apps, new AI video & 3D
Plus Runway Restyle, Luma Ray2 img2vid keyframes & extend
Reading time about 6 minutes.
Welcome to Visually AI!
🔮AI News this Week
Top 100 Generative AI Apps
In the latest edition of Andreessen Horowitz's "Top 100 Gen AI Consumer Apps," the generative AI landscape has undergone significant shifts.
Notably, DeepSeek has emerged as a leading competitor to ChatGPT, while AI video models have advanced from experimental stages to more reliable tools for short clips. Additionally, the rise of "vibecoding" is broadening the scope of AI creators.
The report also introduces the "Brink List," highlighting ten companies poised to enter the top 100 rankings.
ChatGPT's growth trajectory has been remarkable. After an initial plateau, its user base surged from 100 million weekly active users in November 2023 to 400 million by mid-February 2025.
This growth aligns with key product updates, such as GPT-4o's multimodal capabilities, Advanced Voice Mode, and the o1 model series, which enhanced reasoning skills. Mobile adoption has also been strong, with 175 million users now accessing ChatGPT via the app. These developments underscore the dynamic evolution of consumer AI applications.
DeepSeek hit 10 million users within 20 days, compared to ChatGPT achieving the same in 40 days.
AI Video
If you’re a regular Visually AI reader, you have seen many examples of the rapid pace of innovation in AI video models, like Kling, Hailuo, Runway, and more.
Consumers are taking advantage of platforms offering multiple video models, like Krea and Freepik.
AI Coding
There has been significant growth in AI-powered tools for coding in text-to-web apps (Bolt and Lovable) and Agentic IDEs (Cursor).
AI coding tools are often used together and with LLMs such as Claude and ChatGPT, known for advanced coding knowledge, reasoning, and problem-solving abilities.
Read more about other gen AI categories in the report here.
Top 50 Gen AI Web Products
Top 50 Gen AI Mobile Apps
📸 AI Snapshots
Moonvalley Marey
Moonvalley announced the launch of Marey, the first clean AI model trained on fully-licensed data. Marey is the first-of-its kind HD video model, outputting the highest precision AI videography to-date, and production-ready control over cameras, subjects, motion graphics, and more. Sign up on waitlist here.
Intangible
Intangible is the first AI-native 3D platform that eliminates the need for coding and simplifies 3D design with a browser-based tool that combines:
Simple text prompts to generate and manipulate 3D worlds
A proprietary knowledge graph that understands human concepts for 3D
AI agents that automate complex workflows
Intangible was founded by Apple, Pixar, and Unity veterans and just emerged from stealth with $4M in seed funding from a16z Speedrun, Crosslink Capital, and industry angels. Sign up for early access.
Runway Restyle
Runway added a new video Restyle feature that replaces the original first frame of your video with a new first frame - but it has to match the structure of the original frame exactly. You can upload a video and there is an option to download the original first frame, to retexture on Magnific, Midjourney or another tool.
You can use this to Restyle videos from 3D animations, illustrations, black and white, and more.
I’ve been using this feature to change the lighting in videos, and I use the same starting image to prompt for “static hyperspeed timelapse, showing the sky quickly transition to a stunning [midnight / sunny] sky.”
Then, I extract a frame from the new video with a free tool called Video Frame Extractor to use as the Restyle 1st frame in Runway:
Luma Ray2
Luma Ray2 now supports Keyframes, Extend, and Loops for image-to-video. You can extend videos up to 36 seconds total. I generated this 28-second video using keyframes for the initial prompt and continued extending versions:
🚀 My Recent Top AI Tools Picks
Manus: General AI agent that uses real-world tasks through step-by-step replays. Waitlist signups available.
Topaz Project Starlight: Topaz Labs’ diffusion-based video restoration and upscale model. Starlight isn't one of Topaz Video AI's filters - it automatically upscales, denoises, and deblurs video footage without the need to change parameters or settings. You can try it with 3 free 10-second renders per week, for a limited time - sign up here.
Arcane Labs: Streamlines large language model (LLM) application development by enhancing agent memory and facilitating easy integrations with APIs and databases.
Flow Edit: Hugging Face demo for Flow Edit, Inversion-Free Text-Based Editing Using Pre-Trained Flow Models, lets you change or replace objects within your image with a text prompt.
I tested it and changed the kitten in the image below, with a baby bear:
Diffusion Self-Distillation: Hugging Face demo for a new research paper for zero-shot customized image generation to put characters or objects in different poses and settings with a single image and text prompt.
I tested it with this portrait of a woman using this prompt: “This character drinking coffee in an outdoor Parisian cafe”
iterapic: Create photorealistic product shots and lifestyle visuals for social posts in seconds.
🖼️ Image Prompts
A great prompt from my friend Ludovic Creator - just change the subject in [brackets] to your preference:
Prompt: A portrait of a [vintage train] made of words, in the style of vector illustration, golden light, high contrast, silhouette lighting, black background, typography art
![A portrait of a [vintage train] made of words, in the style of vector illustration, golden light, high contrast, silhouette lighting, black background, typography art](https://substackcdn.com/image/fetch/$s_!WBta!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Ffc817ee6-0dda-43dd-b7ed-52461ab55f9b_1456x816.png "A portrait of a [vintage train] made of words, in the style of vector illustration, golden light, high contrast, silhouette lighting, black background, typography art")
I used some helpful keywords shared by Drew Brucker on 𝕏 for unpolished realism effects in AI-generated images. “Candid potography”, “subject cropped awkwardly”, “motion cut-off”, help the AI understand the result shouldn’t be AI-perfect. You can change the subject, setting, and action for your prompts:
Prompt: Candid photography of a [young Latina woman in her early 30's] [shopping in Urban Outfitters], subject cropped awkardly, with motion cut-off
🎥 Video Prompts
I posted tips for Aerial city landscape videos:
Prompt: Slow, expansive aerial orbit around a curved river at sunrise, capturing the warm sunlight gently spilling over the cityscape. Subtle reflections ripple across the water's surface, while moving clouds cast natural shadows on mountains and city buildings, enhancing depth and realism.
Why it works well:
"Slow, expansive aerial orbit" - subtle, cinematic movement around the curved river as a focal point
"Subtle reflections ripple across the water's surface" - realistic water movement captures sky color changes
"Moving clouds cast natural shadows" - natural atmospheric changes to add depth
This example used an image from Midjourney, but you can use the same concept for different aerial landscape views.
Midjourney prompt for image: A majestic aerial view of a winding river encircling a green island at sunset. The river snakes through a cityscape and rolling countryside, with distant mountains providing a dramatic backdrop. The sky is filled with soft pink and blue hues, mirroring on the water’s surface.
Prompt: Low-angle shot from the ground, looking up at a helicopter hovering over a desert. Harsh midday sun creating strong contrasts and shadows. Warm yellows and browns with a clear, bright sky. Warm yellows and browns with a clear, bright sky.
Veo 2 on labs.google (I have early access, but it is available on Freepik, Krea, Nim, Imagine)
Thank you for reading.
Have a creative week!
Hello, thank you very much for another creative menu, a lot of interesting and new, you also creative success and to have enough time for everything