The JSON prompting isn't meant for humans. It serves as a structured, machine-readable format that a language model generates from a simple user prompt. This allows the LLM to handle creative expansion and detailed scene description before the diffusion model generates pixels, enabling finer control.

The quality and vision of an AI-generated video are determined more by the source reference images and videos than by the text prompt itself. Providing a strong visual reference gives the model a clear understanding of taste, style, and desired outcome, acting as a more powerful input than descriptive text alone.

Raw internet videos lack direct textual descriptions. To train a video model, teams must first create synthetic datasets by using VLMs or human labelers to generate detailed captions that precisely describe the visual content.

Synthetic data serves as an efficient first step for training specialized AI, particularly when a larger model teaches a smaller one. However, it is insufficient on its own. The final, crucial stage always requires expensive "human signal"—feedback from subject matter experts—to achieve true performance.

Advanced model training is not just about scraping the web. It's a multi-stage process that starts with massive web data, is refined by human-created examples and ratings (SFT), and is then scaled using reinforcement learning on data generated by the model itself. This synthetic data loop is now a critical component.

The breakthrough performance of Nano Banana wasn't just about massive datasets. The team emphasizes the importance of 'craft'—attention to detail, high-quality data curation, and numerous small design decisions. This human element of quality control is as crucial as model scale.

Rather than optimizing solely for performance on standard industry benchmarks, Ideogram focuses on embedding a subjective quality of "taste" into its models. This requires using human designers for evaluation, as they believe current AI is poor at judging aesthetic nuances, giving them a unique creative edge.

Image models like Google's NanoBanana Pro can now connect to live search to ground their output in real-world facts. This breakthrough allows them to generate dense, text-heavy infographics with coherent, accurate information, a task previously impossible for image models which notoriously struggled with rendering readable text.

Scraping images often yields low-quality results like logos and favicons. A clever workaround is to send the top image candidates to an AI vision model (like Claude Vision). The model can analyze the images and identify the best ones, automating a tedious and subjective cleaning task.