TL; DR - LoRA for producing hoverbikes, with and without pilots, using Z Image Turbo checkpoints. Slight slant towards retro-futuristic design based on 1950s and 1970s car and MC design, but has decent flexibility. "hoverbike" was used in every training caption, so I guess it is a trigger word of sorts.
Trained with OneTrainer using Big Love ZT3 diffusers. Tested to work with the ZIT checkpoints Big Love Zia1 (recommended), Vanilla (Original) ZIT, Into Realism and Facsium. Zia1 is vastly superior, but the others work with some extra patience.
Assumed to work with more ZIT checkpoints to varying degrees. Not compatible with Z Base. I may train on Z Base later if there is interest.
This LoRA took a few months (and a LOT of Buzz and credits) to make. The initial goal was to create hoverbikes that drew from 1950s and 1970s car and motorcycle design, but as the project progressed, I injected some sci-fi models as well.
Vanilla ZIT does not easily produce hoverbikes - it associates the concept too much with motorcycles, which means many of them end up with wheels. And the ones that do not are quite limited in design.
This LoRA fixes a lot of that, although depending on which checkpoint you use, it may still occasionally rely too much on its original "bike" concept. To counteract this, use phrases like "hovering a short distance over the ground" "floating on a subtle anti-gravity field" "flying midair", "angled anti-gravity thrusters firing", etc.
These hoverbikes are not strictly bound to hover just above the ground - they can fly higher as well, if you should want that. Personally I have enjoyed making aerial shots over nature and cities. Some of those renders were in the training dataset, so try that if you want to, ZIT is good at aerials.
Since I spent a ridiculous amount of time, buzz and credits on this, I was hoping to set it to early access and earn back at least a little buzz, but it looks like Civitai took away that option for non-members - we can earn 1 buzz per 10 generations - yay, I will be so rich!
All I can say is that if you happen to have a lot of buzz and like the model, I will be immensely grateful for any amount you can spare. I use the bulk of it to create more LoRAs.
Limitations: Since the focus of the LoRA is hoverbikes, the people may be far from the camera which causes the eyes to be indistinct. If you render at higher resolutions, it helps. You can also mask and inpaint, or search for a workflow with an in-built face fix.
Created from images rendered mainly in Nano Banana 2, with a few Seedream 4.5, Nano Banana Pro and GPT Image renders also included.
I plan to update this LoRA within a month or two - feedback of ideas to add to the dataset is very welcome!
I will also soon release other LoRAs unrelated to this one, so click a "Follow" and you will be notified. Enjoy your ride!