Choosing a Model¶
The single biggest lever for quality, speed, and cost.
Let the App Choose¶
Click Auto-select in the model picker and Storywright picks the best model for each task from what's available. Models marked ✦ Recommended are particularly well-suited for their role.
Use auto-select when: setting up for the first time, after adding a new provider, or when you're not sure what to pick.
Three Roles¶
Storywright uses different models for different tasks. Configure each in Settings → Models or per-story in the workspace.
| Role | What it does | What to look for |
|---|---|---|
| Writing | Generates and revises scene prose | Creativity, voice, prose quality |
| Planning | Creates scene outlines and refines plans | Reasoning ability, structured output |
| Extraction | Summarizes, extracts memories and world state | Fast, cheap, reliable — runs 3–4× per scene |
Tip: The extraction model is your biggest cost lever. Switch it to something fast and cheap — it doesn't need creativity, just reliable instruction-following.
Budget Tiers¶
Free — Local Models¶
Run everything locally with Ollama or LM Studio. No API costs. Quality depends on your hardware — 70B+ models compete with cloud; smaller models are fast but lower quality.
Budget — $0.50–$2.00 per story¶
Affordable cloud models. Good for daily writing and drafting. Save money by using a budget extraction model — it runs most often and doesn't need to be creative.
Premium — $2.00–$5.00 per story¶
Top-tier writing models for the best prose quality. Save money by mixing: premium writing model + budget extraction model.
Recommended Models¶
Google Gemini¶
| Role | Models |
|---|---|
| Writing | Gemini 2.5 Pro |
| Planning | Gemini 2.5 Pro |
| Extraction | Gemini 2.0 Flash |
Anthropic¶
| Role | Models |
|---|---|
| Writing | Claude Sonnet 4, Claude Sonnet 4.5 |
| Planning | Claude Opus 4 |
| Extraction | Claude Haiku 3.5 |
OpenAI¶
| Role | Models |
|---|---|
| Writing | GPT-4.1, GPT-4o |
| Planning | o4-mini, o3 |
| Extraction | GPT-4.1-nano, GPT-4.1-mini |
NanoGPT (open-weight)¶
| Role | Models |
|---|---|
| Writing | Euryale 70B, Magnum 72B, Mistral Small Creative |
| Planning | GLM-5.1 Thinking, Gemini 2.5 Flash |
| Extraction | Gemini Flash, any small fast model |
Local (Ollama / LM Studio)¶
| Role | Models |
|---|---|
| Writing | Llama 3.3 70B, Mistral Large, Qwen 2.5 72B |
| Planning | DeepSeek R1, Qwen 2.5 72B |
| Extraction | Llama 3.2 8B, Phi-3 Mini, Gemma 2 9B |
Tips¶
- Test with a short story first. Generate 3 scenes to evaluate a model before committing to a full project.
- Avoid reasoning/thinking models for extraction — they're slow and expensive for tasks that don't benefit from deep reasoning.
- Per-story overrides let you use a premium model for your main project and a budget model for experiments.
- In demo mode, models are pre-selected for you. Add your own API key to unlock the full model picker.