I’ve been using Gemini for the last six months as a reader/editorial suggester on a speculative fiction novel that I have been making, currently sitting at 250 pages. My workflow focuses on structural critique, character voices, biological realism, pacing, emotional continuity across the document, and narrative consistency. I have never used the model to try and generate prose or erotica, and the few times it has started trying to generate prose for me, I very specifically tell it, “Stop. I am the writer. You are the reader. Do not try to generate prose for me.”
This whole Through the entire document, there are exactly two intimate scenes. Always before, the system evaluated the prose I sent with minimal friction. Occasionally, I would have to delete one or two sentences that tripped the filter, but I was able to either do that safely because the necessary information was already contained in context or send that in a separate message for context. I’m assuming following a recent update, yesterday or the day before, I now get immediate hard refusals on those passages. It’s completely unable to distinguish between genuine fiction editing and prohibited generation now.
Challenging the pre-filter buffer by reiterating the non-gratuitous context occasionally forces the model to engage, but it severely clamps the response depth and breaks the creative workflow.
I think there needs to be be Something in place, some kind of workaround for this. Like I said before, as of October 23rd, 2026, I had no problems with the existing mature content filters. They were occasionally minorly restrictive for my writing, but I’d learned to work around them without losing creative flow. Possible solutions are maybe a classifier nuance for editorials. Train the safety filter to differentiate between a request for generating gratuitous content and user-provided fiction submitted for structural critique.
Possibly an optional Mature fiction/creator tier. an age-verified setting or tier or subscription that relaxes blunt keyword scrubbers for people who want to stay in a creative writing workflow. Perhaps even set it up so it specifically cannot generate content. It can make suggestions. It can say, “Hey, if you’re going for this style of wording, perhaps change this,” still identifying character stability, story continuity, all of everything Gemini usually does, but perhaps with a strict no content generation. It cannot generate prose. It can only suggest edits, possibly.
Another thought is Model Version Pinning. If someone who uses Gemini for creative workflow gets used to working with it in a certain way and then an update changes everything, that scrambles it for days, and getting a flow back is very difficult. I think it would be very helpful if we could pin a previous model and use that, so unannounced fine-tuning updates, safety updates don’t just abruptly break an established writing flow. My last possible suggestion is very similar to the opt-in one I made before, perhaps an optional purchasable add-on to Gemini that loosens the filters, not get rid of them, but loosens them for people who do a lot of creative writing and who don’t want to have to constantly censor their workflow.