The AI’s I use like Claude, Deepseek, Grok, and Chat GPT all write in a away that isn’t actual story writing. They usually summarize thing’s, have character’s be omniscient and know past or future events they shouldn’t know, always talk expectantly, give unnecessary sensory and environmental details and it definitely can’t set up a mystery or write a specific genre like comedy. They keep over clarifying as well what something isn’t or is along with even saying what I said not add directly.
I need AI that are at least capable of writing more humanly and following what you tell it. Idk why there isn’t any actual AI built for writing , story writing not essay or school writing. Most can critic something pretty well but can never implement those fixes they suggest. I usually upload my fic through PDF for the AI read . It gets things of what chapter is what at times wrong or completely be unable to remember a chapter ending or the actual one.
Story writing is a bit difficult in my experience. I had more fun with older models like Mistral Nemo. I feel newer AI models are often way more tuned to fulfill the role as a “helpful assistant” / chatbot, which I think tends to make their style of writing worse. You could also try to use some of those “base models”. They’re not tuned in that way. They also won’t follow instructions, they’re more autocomplete. You’d provide them with something like a word problem, give the first few paragraphs and see where they take it.
And honestly, I don’t think AI is super clever, on a book-author level. It’ll always get the pacing wrong. Push for story tropes like sudden plot twists. Introduce random characters to make something happen. And brush over / summarize other parts which would be interesting to tell in detail.
What could help is an elaborate (strict) process. Something like the computer programming / coding agents do. Make it first come up with a story idea. Make a plan, a todo list of the framework story, side stories and arcs, devise chapter names and a short summary of what needs to happen in those chapters. Write short character cards. And only then feed that plan back to the AI and make it begin writing the actual text.
Have you tried providing some reference stories for context via RAG? That should help ground some of the writing. If you wanna go all out you can fine-tune a model to adapt to the style of writing you’re looking for. There should be some “novel writing assistant” models out there because people are definitely writing and publishing books with AI. Good luck!
I don’t know them any at the moment which is why I’m asking. I am not sure where to start in searching. I’ve tried looking but couldn’t find any that actually did even half of what I wanted. I mainly use Deepseek and Claude as my chapters average around 6,500 words or higher and getting even a terrible draft of that I can edit is better than Gemini or Grok where I get a rushed barely 1,000 word “full” chapter or even a scene which also is longer than their limits.
Ah yeah context and token limits are gonna hold you back. Opus output is more than output large chapters, but you’ll probably get the best results by prompting via API (command line/CLI). You can set max output limits and provide files for RAG. You also might need to upgrade to a plan that can handle marathon sessions better like Claude Max. Claude is also better for creative writing apparently. Good luck!
Novel.ai comes to mind. It’s made for writing. It won’t write for you, but it’s better in some ways than the more general purpose AIs you named. I mean, it still needs editing and human input.
The interface is made for that, too. You can just edit the story like you can’t in typical chatbot interfaces. Pacing and omniscience are still a problem, but it does follow specific instructions better. You have lorebooks for specific details. And a very workable amount of tokens
The problem is the process, not the model.
You can’t start by generating the actual text, it has to be an iterative process of setting up characters, timelines, plots, plans, etc. Then running a first pass for planning chapters, then running passes to generate the prose, then editing, revising, and verifying.
Oh I definitely did all that but these AI can’t do most of what I ask. They’re programmed a specific way so it’s hard to get them to actually follow what you want. I made a skill and Use style and project on Claude but it still writes exactly like it always does and has the usual AI habits and additions I noticed every AI has for writing something.
I am going for a style that might be difficult for AI. Close third-person perspective narration with, single quotes used in Thoughts, dialogue and thoughts attached to a longer sentence or paragraph or having a varied attribution or action beat, no literary, abstract, poetic and flowery writing and over clarification also unnecessary detail about stuff.
AI likes to also be very short and punchy sentences, despite how much filler it gives, it can be weirdly not descriptive for certain things and stuff kinda just …happens at times.
“I made a skill and use Style and project on Claude”
You definitely didn’t “do all that” with a single skill. “That” is not even a single agent.
You need to spend 10x more time setting it up than you do writing with it. The writing should be really fast once you get the architecture in place. Also, if you wrote the skill yourself, you’re doing that wrong too.
I would suggest two models: a local gemma 4 26b variant or a cydonia variant. This one specific custom gemma variant I have been using does a very good job at providing vivid details, while not being overly verbose. I love Deepseek 4 but I can assure you I agree with you 100% in that I don’t think it can write “like a human”. I’ve also got some custom versions of GLM4.7 that support very high memory context that run well on consumer hardware and also seem to have very good writing styles. I’ve also used various GLM 5.1 instances in remote TEEs are also good at writing, world building, etc. There’s plenty of other suggestions I’m sure others will give you, such as Qwen variants.
However, model suggestions out of the way:
Writing styles are always going to be tough to get right, regardless of the model.
