So whether you batch or process individually its based on the number of images.
If you take a look at the billing page where it details which API calls are counted as an image and under which scenarios 2 images count as a single request: https://aws.amazon.com/rekognition/faqs/#Billing
For the speed and accuracy I don't think the costs is too high.
You could try making a video from an image batch first using ffmpeg:
ffmpeg -r 1 -i img%01d.png -vcodec mpeg4 -y movie.mp4
Then you could process the video using Rekognition, the pricing is $0.10 per minute of video.
To be cheaper than per image processing you would have to put more than 125 images per minute though.
Amazon Rekognition errorasked 8 months ago
Line over text in Rekognitionasked 10 months ago
Using Rekognition to process a batch of images on an S3 bucketasked 5 months ago
When using Rekognition, if you previously saved 1 million images in a Collection and do a SearchFaceByImage of 1 new image against those 1 million images, do you need to pay for 1 million API calls?asked 4 months ago
Costs for Batch Rekognition OCRAccepted Answerasked a year ago
How to load annotation from local computer in Amazon rekognition custom label serviceasked 6 months ago
Can I use Amazon Rekognition to find cropped versions of an image?asked 4 months ago
Understanding Amazon Rekognition Training which ignores images with WARNING_NO_ANNOTATIONasked 9 months ago
Is one supposed to grayscale and brightness contrast process the image before sending to textract?asked a year ago
Using DetectText in rekognition to extract text from images of low quality.asked 4 months ago