Web applications
ImagiForge
Stable Diffusion XL behind a prompt box, with no account step.
- Role
- Sole author
- Period
- 2025—2026
- Status
- Live
The problem
Most image generators put a sign-up form, a credit balance or a subscription between the idea and the image. For somebody who wants to try a single prompt, that is most of the cost of the exercise. Removing the account, though, removes the thing abuse is normally attached to.
The approach
A Next.js front end calls a private Hugging Face Space running SDXL on ZeroGPU. Rate limiting is keyed on a salted hash of the visitor's IP rather than an account, so the daily limit works without anyone signing in, and the model token never reaches the browser.
What it does
SDXL at native resolution, on GPU
The full base model at 1024px rather than a distilled stand-in. Roughly ten seconds an image on a ZeroGPU Space; the same render takes over an hour on the free CPU tier.
Throttled without accounts
Five images per visitor per day, counted in Redis against a salted hash of the IP. The raw address is never stored, and a global daily cap protects the GPU budget behind it.
Moderation in depth
A word filter runs before any GPU time is spent, then again inside the Space, where a text classifier judges the prompt's meaning and an image classifier checks the finished picture. Term pairing keeps ordinary prompts working — 'a child in a meadow' and 'a mosque at sunset' both generate.
Nothing kept
Prompts are discarded after each request and images are never written to disk. There is no gallery of user output and no analytics.