Nodaro DocsResearch
Studies from the Nodaro team on how AI models really respond to prompts, with the exact wording we tested, the failed takes and the results we shipped.
Nodaro Research is where we publish what we learn while making AI models do what our users ask. Every picker, preset and default in Nodaro is tuned against real generations. These articles show that work: the wording we tried, the takes that failed, the numbers, and the rules we kept.
Each study is a product engineering study, not a statistical benchmark. It names the models, resolutions and conditions it covered, so you can judge how far its findings travel.
Studies
Vision Model Bake-off
Seven AI vision models read the same four portraits at 20 settings. Which setting became the default, why more reasoning only made reads slower, and why the GPT models never named an ethnicity.
Image Model Bake-off
Seven AI image models on 18 real jobs: edits, characters, exact text and presets. Which model won each job, which one won nothing, and all 112 images.
Transition Lab
What more than 250 videos taught us about AI video transitions. 82 transitions, why video models blend by default, and why the start and end frames decide as much as the words.
Camera Motion Lab
It took 147 videos to learn how to move a camera in an AI video model. 64 camera movements, 13 purpose-built base images, and why a start frame can outweigh the prompt.
Frequently asked questions
What is Nodaro Research?
It is where the Nodaro team publishes what it learns while tuning the product against real AI models, with the prompts, the numbers and the clips from each study.
Are these studies benchmarks?
No. They are product engineering studies. Each one finds wording and settings that work reliably enough to ship, and says which models and conditions it covered.
Related
Camera Motion Lab: how 147 videos taught us to move a camera
Camera Motion
Choosing a model
Last updated on