Just some insights as I work into this. While you can use AI to put the car just about anywhere, it feels wrong. The video looks much more natural, when it shows the dealership and original location. You can do the craziest of things, make the car drive in the city, or any sort of thing. But then is just becomes so clear is AI. I think the key is to get it to a point where there is just no way to know its AI generated. Im not sure if that is possible with the tools and models currently available, but is not too far away either. On the previous tests I used Google Veo 3.1 : Gemini AI video generator powered by Veo 3.1 , that is supposedly the top model now as from the leaderboard here : Image-to-Video Arena | LMArena . Now new models come all the time. There is one that is particularly good from OpenAI, and generates much more natural looking clips, yet it does not allow to generate videos starting from pictures, so we cant use it. That said, I think is a certainty that Google and other AI Labs will match and surpass OpenAIs model in the next few months. From my testing it will seem that we are one or two generations away from the videos being indistigisible from real.
AI Video Marketing: Is this the future for dealership sales and service?
- Thread starter playersclubcars
- Start date
Thank you! been improving the code, managed to fix a few things from the previous examples, will post new ones soon.
The results may not be because of the prompts, but simply because Google Veo 3.1 is really good, major leap from previous models. The prompts include details about the cars, the scene, the character and the camera motion. I do not write the prompts but have an LLM write them for me, taking into account all these details.
A few more tests, these are from older code, with some errors on the narrations and video, but let me share them before showing you the new version so there is some contrast:
New code version is much improved, trying all sorts of tricks to push things, I think Im reaching the limits of what the underlaying video model can do.
New code version is much improved, trying all sorts of tricks to push things, I think Im reaching the limits of what the underlaying video model can do.
One more, notice the errors in the badges, source pictures where from a random post on FB so very low resolution. I think with great pictures as source; the badges and other text details will show correctly.
Going to continue working on this code over the weekend, already managed to get a bit more details on the videos. Have a few more ideas to try. They are starting to get acceptable, narration is a bit better too:
Can anyone upload some great quality pictures I could use to generate a video? The more details, the less compression, the better. It needs to be the usual dealership location pictures, just super high quality. Will appreciate the help.
Going to keep improving the code over the weekend, a few updates already on the current version. Still seeing some errors from time to time. Very obvious on this one if you pay attention:
Source pictures came from this facebook group: Exotic Cars For Sale By Owner
Will start a new thread soon with all the details on how to make these.
Source pictures came from this facebook group: Exotic Cars For Sale By Owner
Will start a new thread soon with all the details on how to make these.
- Jan 31, 2024
- 8
- 0
- First Name
- Md Abul kalam
That’s a really smart approach — cutting down production time while keeping the personal touch is a huge win for dealerships. AI video like this can make communication faster, more consistent, and still feel genuine to customers. The G63 AMG example sounds like a great showcase of what’s possible. Excited to see how this tech continues to evolve!
