Why we built this library
Before we put this collection together, we had an internal debate: where do users actually get stuck on the platform?
The answer wasn't "they don't know how to use it" — it was they don't know if it will work. Can a given model actually produce a white-background product image that meets e-commerce listing standards? Does Seedance 2.0's "physics simulation" actually hold up for food product videos? Is Veo 3.1 worth the price for commercial ads?
These 35 cases are our answer to those questions.
How we tested
The 35 cases are spread across five model families (NB1, NB2, Pro, Veo 3.1, Seedance 1.0/1.5/2.0) and cover more than ten industries including e-commerce, food service, fashion, beauty, home goods, and education. For each case we recorded:
Every image and video was generated as-is, without post-processing. Some carry the model's typical flaws — we think seeing failure cases is more useful than a highlight reel.
- The complete prompt (copy and use directly)
- Parameter settings (aspect ratio, duration, reference images)
- An honest description of the output, including failures
- A practical business judgment ("can this image go to a client right now?")
A few findings worth noting
Nano Banana Pro performed better than expected on premium product photography. We expected Pro to shine mainly on creative/conceptual requests, but it handled complex materials (leather, metal, glass) far better than NB2. Luxury-tier product shots needed almost no iteration. The trade-off is longer generation time and higher cost.
Seedance 2.0's multi-shot capability is real, but it needs precise prompts. We tested food, beauty, and fashion categories — with very different results. If you don't explicitly describe where cuts should happen, the model decides on its own, and the rhythm rarely matches commercial editing conventions.
Veo 3.1's native audio is harder to control than we expected. There's still a noticeable gap between describing a sound in text ("soft piano") and what actually gets generated. If you have specific audio requirements, we recommend generating the video first and adding audio in post.
The library will keep growing
35 is just the starting point. We keep adding cases, focusing on the scenarios that come up most often in user feedback. Browse the case library to see the full set with copyable prompts — and if there's a specific industry or use case you'd like to see, tell us through the contact page.
One note on availability: on Nano Banana, image generation runs on Nano Banana 2 Lite in production, and the video workbench is currently in demo mode (placeholder results, no credits charged). The video cases in the library document what each model can do, so you're ready the moment real rendering lands.