I'd say, drop the idea that the filter is supposed to be perfect. A filter that is little-used can get a rough content first time around, preferably specified by the person asking for the filter, then people using the filter can suggest adding or removing images. Volunteers can go and work on the filters if they want, but if they don't, the filter will just be changed by such suggestions.
Then again, there is the alternative of only including filters with at least a certain amount of expected usage. I see no problem with not having a filter for everyone who asks for it. I don't think that doing things perfectly and not doing them at all are the only options.
I don't except it to work perfectly. Nothing is perfect by default. But even if it would perfectly we provide a simple tool (the filter labels/categories) to censors, to improve their doing, while we, the volunteers, would indirectly support them in doing so.
For example: The head of a group (state, religions group, ...) of people is trying to censor Wikipedia, because it might damage it's position. What would be easier to comply at the mailing list that a filter for xyz is seriously needed. Now he can start to add images to this filter, calling for volunteers that have to obey to do so. At the end we represent the opinion from the head of the group (not the individuals, that fear the head), publish it as consent and help them to justify their position.
What would someone living inside such a group think if the content is already labeled that way, that he should not look at it. Isn't it social pressure put on the free mind, especially if other members of the group are around?