This week on Denoised, hosts Addy Ghani and Joey Daoud cover the latest wave of AI image generation updates from major players, examine why Hollywood productions are increasingly filming overseas, and explore how Roblox is bringing generative AI to 3D object creation. With multiple new releases competing for attention in the AI space and production economics shifting dramatically, there's plenty to unpack in this information-packed episode.
The AI Image Generation Race Intensifies
The past week has seen a flurry of activity in the AI image generation space, with multiple companies releasing significant updates to their models and capabilities. This acceleration highlights how quickly the technology is advancing and how competitive the field has become.
Reve Makes a Splash with Photorealistic Output
Joey kicked off the discussion by highlighting Reve, a new image generation model that emerged early in the week. Despite being a newcomer, Reve immediately impressed with:
Highly accurate, photorealistic images from their own proprietary model
Exceptional text handling capabilities, an area that has traditionally been challenging for AI image generators
A preview beta that showcases the model's impressive output quality
Though details about Reve's training data remain undisclosed, the quality of its output puts it immediately in competition with established players. Addy noted that this represents another American company entering the increasingly crowded AI image generation market.
ChatGPT 4o Brings Powerful Image Generation to Mainstream Users
The biggest news of the week came when OpenAI announced a major image generation update to their ChatGPT 4o model during a livestream with CEO Sam Altman and developers.
This update represents a significant advancement because:
It integrates high-quality image generation directly into the familiar chat interface
Users can create and modify images through natural language without needing technical knowledge
The system demonstrates impressive spatial understanding
Joey highlighted several examples that showcase the model's capabilities:
One user uploaded a 3D model of a dragon and requested a wireframe overlay, which the AI accurately mapped, demonstrating its understanding of 3D spatial relationships in 2D images
Users can easily modify rooms or spaces by uploading a photo and requesting specific changes
The system can transform hand-drawn sketches into photorealistic images while maintaining the original composition
Addy noted this represents a significant shift in accessibility: "This is like really advanced control net stuff where you can guide the inference and you're doing it with just basic prompting." The integration eliminates the need for complex UIs or technical knowledge that previous systems required.
The model also includes important safeguards against creating deepfakes or manipulating images of real people, a crucial feature for responsible AI deployment.
Ideogram 3.0 Focuses on Design and Brand Applications
Completing the week's trilogy of major releases, Ideogram launched version 3.0 of their image generation model. Joey has been a fan of their previous versions, particularly for their text handling capabilities.
The new update focuses on:
Enhanced style adherence across different images
Expanded capabilities for creative brand graphics and design work
Improved tools for logo creation and graphic design applications
Ideogram's interface continues to be a strength, offering users guided options beyond just text prompts, such as buttons to specify whether you're aiming for realistic images, animated styles, or graphic designs.


