Generate without exposure.
Private image infrastructure for developers. Run leading open-weight models straight from your terminal — encrypted in transit, processed on ephemeral compute, and deleted the moment your file is delivered.
Open weights · Uncensored · Zero retention · Pay per image
Open weights
Pinned by digest — never silently upgraded.
Zero retention
Prompts and outputs deleted after delivery.
Pay per image
Charged only on successful generations.
Ephemeral compute
Jobs run on short-lived, isolated workers.
The privacy of local inference. The convenience of cloud compute.
Unexposed gives engineers local-grade control over image models — without operating the GPU stack.
Private by architecture
Encrypted queues, ephemeral compute and no hosted gallery. Every privacy claim maps to a control you can verify — not a slogan.
Raw, uncensored output
No content filter, no silent prompt rewriting, no hidden negatives, no injected house style. The model you pin is the model that runs — complete creative control over lawful work.
Built for automation
CLI and API, batch jobs, cost caps and deterministic seeds. Image generation that drops into pipelines, agents and CI.
No GPU ops
No CUDA, drivers, model downloads or queue tuning. Send a request, receive a file written straight to disk.
$ unexposed modelsflux-dev text→image $0.08 open-weightsdxl text→image $0.04 open-weightqwen-image text→image $0.06 open-weight$ unexposed gen "editorial product photo of a watch" \--model flux-dev --seed 7 --out watch.png✓ written to ./watch.png · retained: 0 bytes
Choose a model. Run a command. Receive a file.
Choose a model
One command — unexposed models — lists every open-weight model with its price, digest and capabilities. Inspect one before you commit.
Run a command
One explicit command. Pin the model, seed and digest for generations you can reproduce, byte-for-byte, later.
Receive a local file
Your image streams to the client and lands on disk. No hosted gallery, no public URL, nothing left in our cloud.
No moderation layer. Your prompt, your call.
Lawful work runs unfiltered — no content filter, no silent rewriting, no house style imposed on your output. The only boundary is the law, never our taste.
Your content exists only long enough to create your image.
Every privacy claim is paired with an architectural control. The prompt enters an encrypted queue, runs on short-lived compute, and is removed once your image is delivered.
Client
Prompt + source file
Ephemeral queue
Encrypted, content-free metadata
Isolated GPU
Model pinned by digest
Streamed output
Written to your client
Memory cleared
No hosted gallery
No prompt history
Default product behaviour
No content in logs
Documented logging schema
No public output URL
Output streams to the client
No third-party proxy
Models run inside our boundary
We don't claim "private even from us." We claim exactly what the architecture enforces — and we show you the controls behind every line.
A transparent catalogue, not an opaque provider chain.
Every model lists its version, price, capabilities and digest before you run. The model you pin is the model that runs — no silent substitutions.
Pay for images, not access.
Usage-based and legible. The price is quoted before the run, and you are charged only when an image is delivered.
- See the exact price before every run.
- Pay only for images that are successfully delivered.
- Buy credits with no subscription required.
- Purchased credits don't expire arbitrarily.
- Set per-request and account-level spend caps.
Estimate before run
- Model
- flux-dev
- Resolution
- 1536 × 1024
- Images
- 4
- Max cost cap
- $0.50
- Estimated total
- $0.32
# charged only on successful delivery
Your prompt. Your model. Nobody watching.
Join early access and generate from your terminal in minutes. Encrypted in transit, run on ephemeral compute, deleted after delivery.