I was generating videos within google ai studio, with following method:
generate a video.
after examining it, download the video and make a capture of last frame.
feed the last frame as reference image, generate next video.
I generated few video until like ~8 hours before, but suddenly, video generation failed.
after that, I tried with same method several times with veo studio and using gemini api, got following errors:
{"error":{"code":400,"message":"The string value `…` for `bytesBase64Encoded` exceeds the length limit. Please adjust your request accordingly.","status":"INVALID_ARGUMENT"}}
in `…` there there was a long string, ImageBytes object
I tried to make file size small, about 20 kb it could generate a video without problem. but I can confirm I generated the video with ~1mb png with 1280x720 dimensions. is this change intended? is this change documented? is this a right place to ask this?
Same issue, no longer being able to generate any video.
For context, I have Google Cloud Run-deployed application, created by AIStudio, that send API requests incl. one or two reference images to the Veo API and returns a new video.
The response is same to yours
{“error”:{“code”:400,“message”:“The string value `…` for `bytesBase64Encoded` exceeds the length limit. Please adjust your request accordingly.”,“status”:“INVALID_ARGUMENT”}}
Looks like the issue is on Gemini API for all Veo model but not sure if this happen to everyone.
And not just reference image for me, but first frame and last frame image as well.
Tried with Vertex AI with the same model and the error does not happen.
Waiting for more info
Having the same (on veo-3.1-generate-preview), and I whish that this was documented somewhere (in case something indeed change, and it’s not just a temporary issue).
Returning the entire base64 image in the error is probably not intended.
Hey folks! Can you share the aspect ratio and size of the reference image used to generate the video? And if possible the image itself? I tried generating a video with a few different references images and I am not able to reproduce this issue.
As I just tried, I could generate video with ~800kb image, in both google ai studio and using gemini api. so I think the issue is gone(at least for me).