What are this project's goals? How will you achieve them?

Noisify protects personal photos from AI nudification before harm happens. A user uploads a photo, picks a protection level, and gets back a version with invisible adversarial noise. If someone puts that photo into an AI undressing tool, the output comes out destroyed: distorted, unusable, unrecognizable.

There are zero consumer tools that do this today. More than 90 nudification services exist right now, many of them free, most giving results in under a minute. 96% of deepfakes are made without consent, 99% of sexual deepfakes target women. Legal tools kick in after the harm has already spread. Noisify works before anything happens.

The goal for the beta is documented, measurable degradation of output quality across the three most used nudification model families (SD 1.5, SDXL, Flux), tested and published honestly.

How will this funding be used?

We are asking for $10,000 minimum and $25,000 ideally, for a 6-month runway to a working beta.

Minimum ($10,000): GPU compute on-demand ($1,800), infrastructure and tooling ($700), partial team stipends ($7,500).

Ideal ($25,000): GPU compute ($2,000), infrastructure ($1,000), part-time compensation for 3 people at around 10 hours a week each ($21,000), security audit of the inference pipeline ($1,000).

The plan for self-sufficiency is small per-use fees to cover compute once the beta proves efficacy.

Who is on your team? What's your track record on similar projects?

I'm Vlada, the founder. My background is in environmental engineering, physics, and art-science practice. I handle product decisions, research, and documentation. Academic research is probably my strongest skill.

Anton is our backend and ML engineer. He currently builds AI infrastructure for industrial clients and is writing the Python inference layer and Go backend, and putting together the model ensemble.

Nikita is an information security specialist and Python/C++ developer, handling infrastructure and secure architecture.

In two weeks of part-time work we have written technical research documentation on SD 1.5 and SDXL model families used in nudification, set up S3 storage with model weights on-demand, created a GitHub organization, and written the first inference code. This week the goal is the first protected photo output.

What are the most likely causes and outcomes if this project fails?

The main technical risk is that adversarial perturbations get stripped by preprocessing in closed commercial pipelines: compression, downscaling, color normalization. We are working around this by targeting low-frequency image components where possible, which are much harder to remove than high-frequency approaches like PhotoGuard. We are not going to overclaim. No tool today gives 100% protection and we will document what works and what doesn't honestly.

If we don't reach sufficient efficacy, the outcome is still a public technical report on what we tried and learned, which is useful for anyone else working on this problem.

How much money have you raised in the last 12 months, and from where?

No external funding so far. The team has been working voluntarily. We are currently applying to grantmaking.ai and Emergent Ventures in parallel.