𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗚𝗲𝗺𝗶𝗻𝗶 𝗢𝗺𝗻𝗶 𝗙𝗹𝗮𝘀𝗵 𝗳𝗼𝗿 𝗩𝗶𝗱𝗲𝗼… 𝗮𝗻𝗱 𝗜 𝗵𝗮𝗱 𝘁𝗼 𝗳𝗶𝗴𝗵𝘁 𝘁𝗵𝗲 𝗴𝘂𝗮𝗿𝗱𝗿𝗮𝗶𝗹𝘀 𝗳𝗶𝗿𝘀𝘁 😅
Personally, I’m usually less interested in pure text-to-video models.
What caught my attention with
@googledeepmind Gemini Omni Flash was all the video editing and V2V examples people were sharing, so of course I had to test it myself with SKY.
But wow… getting past the “this generation might violate our policies” warnings was a challenge.
I tried:
• My own 2D generated animations → blocked
• Stock footage of an old guy skateboarding → blocked
• Different source videos → blocked again
Eventually I created the original source clip inside the same workflow/platform… and finally it worked 😂
Honestly, I hope these guardrails become less aggressive in the future because right now they make human-based video editing unnecessarily difficult.
That said… once I got past that part, I couldn’t stop experimenting.
Most of these videos were generated with very simple prompts. For a few, I also used reference images, although the model definitely won’t follow your style reference 100%.
Some quick thoughts after testing:
• Prompt following is usually good (never perfect! I did one comparison of V2V with seedance and Gemini did better)
• Generation speed is very fast compared to many other models
• Mixing 2D/3D aesthetics with realism is incredibly fun
• The model feels flexible and playful for creative experimentation
Downsides:
• Resolution is limited to 720p
• Cleaner/sharper detail still isn’t on the level of Seedance or Kling in my experience
• Style consistency with reference images can drift quite a lot.
• Still Changes the character a bit (it did mess up SKY's hair color a lot)
Still… overall I’d honestly describe it as kind of a “Nanobanana for video” 😄
I added some of the prompts directly into the video, but if you want the full prompts, comment “prompt” and I’ll send them over 👇
Oh, and btw… the full Bananas song is coming soon
