Field notes
AI filmmaking, tested in production
Practical guides, model comparisons, and production lessons from building cohesive films with generative video.
Featured · AI writing modelsWe had 8 LLMs write 104 screenplay scenes. Results varied by creative directionA controlled, blinded OpenRouter experiment comparing Claude Opus 5, GPT-5.6 Sol, Kimi K3, DeepSeek, Nemotron, Inkling, Gemini, and Qwen on the same dramatic premise under five creative directions. Every script is published.Read the analysis →Latest articles
5 articlesAI video modelsWe replaced two dancers with AI characters: Wan 3.0, Kling, MiniMax, Omni and AlephOne dance video, two fictional portraits, five video-editing models. Watch the source and outputs, compare measured costs and resolution, and see where character consistency and moderation broke the workflow.AI image modelsWords, a reference image, or an AI description? We ran 252 tests and changed our answerFive blinded image-generation experiments across stylized 3D, cut-paper, gouache, linocut, and stained glass. The method that won the first experiment lost the next four, with practical takeaways and limits for choosing your own prompting method.AI video modelsMaking AI storyboard sheets reliable: five rounds of experiments with Seedream 4.5Our landscape test passed 29 of 30 sheets after prompt iteration. Vertical tests on two models exposed layout failures; panel detection recovered useful options, with important limits.AI video modelsHow we compare AI video models for filmmakingA proposed rubric for judging video models for narrative work. Character consistency, direction-following, temporal coherence, style stability, and the number that matters most, cost per usable shot.