I also wonder... Even if the graphics are made with just the character set, are they not a bitmap graphic nonetheless? After all those characters are just bits mapped in a cell (8x8 or other depending on the system). So no modification to those stock character then? Maybe the OP should have asked for pure ASCII art then?
As far as I know, in graphics mode 1 you can redefine the character set with 256 symbols that you position freely on the screen and which can be repeated any number of times. If you would make a pseudo-bitmap screen, it would be 16x16 characters, 24x10 characters, 32x8 characters or something similar, while the total screen area still is 32x24 positions. Furthermore, these 256 symbols are divided into 32 colour sets with 8 symbols in each. Each colour set has its own combination of background and foreground colour, meaning that if you want both a blue ball on white background and a red ball on white background, you would need to define the ball twice in your character set and position those into different colour sets.
A real bitmap in graphics mode 2 takes up three character sets so you can get all 32x24 = 768 positions with unique patterns in each character. The colour tables also are arranged in such way that for each position, you have 8 lines of colour data: one background nybble and one foreground nybble meaning you can have two colours per line and character.
The real challenge then is to pick up to 256 custom symbols arranged in 32 sets of colour combinations, and make it look like a true bitmap image where each position on the screen has its own definition and own set of colours. Obviously the colour resolution in graphics mode 1 is nowhere near the bitmapped mode, but then again not all bitmap images utilize the colour resolution.
As a middle way, some kind of ASCII/ANSI/PETSCII/ATASCII art could be considered. It would involve a custom but fixed character set that probably is duplicated a number of times within the character generator to overcome the colour set limitation unless you want mostly monochrome images.