Upload your product shots, scenes, and lighting once. Mention them in any prompt as @aliases — across Google and OpenAI's best image models, with your whole team.
Model
References4/14
Prompt
Why Diffusion State
Alias-driven prompts
Mention @product twenty times — it attaches once. Unknown aliases are caught before you spend a cent.
Reusable reference library
Originals, optimized provider inputs, and thumbnails managed per project, stored privately in your own bucket.
Two providers, one prompt
The same prompt and references compile natively for Nano Banana Pro (Gemini & Vertex) and GPT Image 2.
Bring your own keys
Org-level credentials, envelope-encrypted with AES-256-GCM. Never sent to the browser, never logged.
Reproducible history
Every job stores the original prompt, compiled prompt, reference manifest, settings, and usage — rerun any of it.
Real multi-tenancy
Workspaces, roles, invitations, audit logs — with Postgres row-level security enforcing isolation.
How it works
Upload references
Drop in product shots, scenes, lighting refs. Each gets an @alias your whole team can use.
Write one prompt
"Place @product in @scene, use @lighting only for lighting." Aliases highlight as you type.
Generate anywhere
Pick Nano Banana Pro or GPT Image 2. Same prompt, native compilation, outputs in your library.
Security
Diffusion State is architected as a real multi-tenant SaaS: every record belongs to a workspace, and Postgres row-level security enforces isolation below the application layer.
Illustrative numbers — every workspace gets its own dashboard.
Free to set up — you only pay your own OpenAI and Google usage.
Create your workspace