Fire
Our flagship service, a 7b-parameter text model with some contextual enrichment to refine outputs. Fire is a top-tier instructional model that feels the most like using a raw LLM, with just enough app-level functionality to reduce noise and improve productivity.
Model Details
A high-quality, instructional text model
7b-parameter text-to-text LLM
Extensible and flexible
32k token context
Infinite context-isolated sessions
1 credit per prompt† (see Pricing)
Mirage
Our first text-to-image model, based on Stable Diffusion with added app-level session history to enable iterative prompt extension. Now you can prompt between results to narrow down the exact artwork you envision.
Model Details
A high-quality text-to-image model
Based on Serenity (a SD 1.5 fine-tune)
Iterative prompt extension capabilities for ongoing creation
512x768 (portrait) and 1024x768 (landscape)
Infinite context-isolated sessions
2–4 credits per image (see Pricing)
Flow
A 7b-parameter text model with added retrieval-augmented generation (RAG) capabilities to enable domain-specific responses with high accuracy. Flow is an advanced RAG-based approach that pre-computes the LLM context from files (user uploaded or otherwise) resulting in precise, extensible knowledge.
Model Details
Advanced text-to-text LLM with RAG
Dynamically extend context with file uploads
Upload up to 10GB of data
Infinite context-isolated sessions
2 credits per prompt†
2 credits per file upload†† (see Pricing)
Coming soon: Cerulean
Our first text-to-video model, based on Stable Video 4D (SV4D) with added app-level session history to enable iterative prompt extension. Prompt between results to narrow down the exact shot you envision.
Model Details
A high-quality text-to-video model
Based on Stable Video 4D (SV4D)
Iterative prompt extension capabilities for ongoing creation
960x540 and 1920x1080 (HD)
Infinite context-isolated sessions
4–8 credits per 1 second of video (see Pricing)