A Matlab/Octave program that generates all possible gray images for a given image size.

allGrayImages.m

clc
clear all
close all
format long

imageProperties.height = 2;
imageProperties.width = 2;
output_folder = 'image_output';

% only used for estimations, not for the actual image calculation
imageProperties.bits = 4; 
size_of_single_image_in_bytes = 90;

maximumPixelValue = (2 ^ imageProperties.bits) - 1;

estimations.imageCount = (maximumPixelValue + 1) ^ (imageProperties.width * imageProperties.height);
estimations.totalSize.bytes = estimations.imageCount * size_of_single_image_in_bytes;
estimations.totalSize.kb = estimations.totalSize.bytes / 1024;
estimations.totalSize.mb = estimations.totalSize.kb / 1024;
estimations.totalSize.gb = estimations.totalSize.mb / 1024;
estimations.totalSize.tb = estimations.totalSize.gb / 1024;


disp('All gray images:')

formatString = ['%0' num2str(imageProperties.width * imageProperties.height) 'x'];

% use parfor if you have the parallel computing toolbox installed
for image_id = 0:(estimations.imageCount - 1)
% parfor image_id = 0:(estimations.imageCount - 1)
    image = zeros(imageProperties.height, imageProperties.width);
    imageContents = sprintf(formatString, image_id);
    
    imageIsInteresting = true;
    firstValue = imageContents(1, 1);
    
    imageContents = reshape(imageContents, imageProperties.height, imageProperties.width);
    for col = 1:imageProperties.width
        for row = 1:imageProperties.height
            image(row, col) = 255 * hex2dec(imageContents(row, col)) / 15;
            if ~imageIsInteresting && imageContents(row, col) ~= firstValue
                imageIsInteresting = false;
            end
        end
    end
    
    if imageIsInteresting
        disp(num2str(image_id))
    end
    imwrite(uint8(image), [output_folder '/' num2str(image_id) '.png'])
end

Beware: The amount of generated data is HUGE!

To keep the computation simple and fast, the values a pixel can take only range from 0 to 15 (scaled to 255 for visualization). This makes the computation easy as it allows each image to be represented by a 4-digit hexadecimal number.