Summary
Dan Kieft presents a full workflow for creating realistic AI images and video using the Higgsfield platform, which sponsored the video. He tackles the most common question, which model to use, and his answer is that it depends on what you are generating.
The guide compares the same prompts across several image models, including GPT Image 2.5, Flux 2.0 Pro, Flux 3, Nano Banana Pro and 2.1, Seedream 5.0, and Higgsfield Soul 2. Flux 2.0 Pro still stands out for realism, while GPT Image 2.5 impressed him on several prompts and some models struggle with text rendering such as timestamps. His advice is to generate with many models at once since images are cheap.
The video then moves to AI video, comparing Seedance 2.5 at 1080p and in 480p draft mode, Minimax Hailuo 3, and others. It notes that lower-quality draft mode can help produce a realistic, less polished look. Dan also shares a prompt formula covering camera, tone, subject and action, lighting, and texture, plus an LLM-assisted prompting template. Creators and filmmakers get practical tips for achieving lifelike footage without needing a film background.
📺 Source: Dan Kieft · Published October 16, 2026
🏷️ Format: Tutorial Demo
