Five providers in the box. No credits, ever.
Re-Prompter has no inference bill, because it never touches your tokens. Your prompt goes from your browser to the account you already pay for, or to a model running on your own machine.
What each one is good for
Nobody is forced to do everything.
| Provider | Text | Vision | Image generation | Image editing |
|---|---|---|---|---|
| OpenAI | yes | yes | yes | yes, with masks |
| OpenRouter | yes | model-dependent | some models | some models |
| Anthropic Claude | yes | yes | no | no |
| Google Gemini | yes | yes | image-capable models | image-capable models |
| Ollama (local) | yes | vision models only | no | no |
| OpenAI-compatible | yes | you declare it | you declare it | you declare it |
Claude does not draw, and Re-Prompter never pretends otherwise. It writes the briefs, reads the references and interprets the edits; pair it with an image provider and the router handles the rest.
Routing
Eight tasks, each with a preferred and a fallback.
Brief writing, vision analysis, image generation, image editing, face swap, title generation, style extraction and persona analysis. Press Route automatically for a sensible default the moment you connect a provider, then override anything.
Test connection
Every provider has one, and it tells you something useful: how many models the key can reach, how many of them return images, how much of an OpenRouter budget is spent, how many Ollama models are installed and how many read images.
Searchable model lists
Pulled live from the provider, with capability badges and, where published, the price per million tokens. Or type a model identifier yourself.
Ollama, properly
Detect the server, list what is installed, and ask each model's own manifest whether it has a vision projector — so a text-only model is never handed an image and left silent.
Anything OpenAI-shaped
LM Studio, vLLM, LocalAI, llama.cpp's server, a self-hosted gateway. Name, base URL, key, model, capabilities. No code change and no update needed.