AI Policy · Duskflash

Duskflash is a film and digicam camera first. AI is an optional part of Pro, and it has one rule above all others: it must not change who you are. This policy explains how AI Studio and our AI assist features work, which companies process your data, how we keep results safe, and what is not allowed.

1. What uses AI, and what does not

No AI involved: every film, digicam, disposable, instant and black-and-white look, the flash simulation, grain, halation, light leaks, date stamps, frames, the photobooth and film rolls. These are deterministic image processing on your phone. They never redraw your face and never upload your photo.

AI Studio (Pro): styles such as Flash Relight, Golden Hour, Night Out Flash, Vacation Scene, Photobooth Studio and Restore Old Photo. These send the photo you choose to an AI image model, which relights or re-scenes it.

AI assist (Pro): Describe a look (turns your words into a new local look), captions and hashtags, the photo-dump planner and look suggestions. These send text only to DeepSeek, never images.

2. Before anything is uploaded

The first time you open AI Studio, the app explains which providers are used and how long data is kept, and asks for your consent. You can withdraw consent in Settings → Privacy → AI at any time.
Your phone checks the photo first (a face is visible, it is sharp enough, bright enough and large enough) so you don't spend credits on a photo that will not work.
The photo is resized to at most 2048 pixels on the long edge and its metadata, including location, is removed.
Free-text image prompts are not offered. Styles use prompt templates written and reviewed by our team.

3. Identity Lock

Identity Lock is how we keep AI results looking like you.

Faithful prompts. Every style tells the model to preserve the person's identity, facial features, skin tone, hair, body shape and expression, and to change only light, color and scene.
On-device compositing. When the result comes back, your phone aligns it with your original photo and blends your real face back in through a soft mask, matching tone and light.
On-device likeness check. Your phone then compares each result with the original and shows a likeness score (for example "97% you"). Results below our threshold are hidden and their credits are refunded automatically.
Your call. If a result does not feel like you, tap Not me and the credit comes back.

No biometric data leaves your phone. The face landmarks and numeric face signature used for steps 2 and 3 are computed on your phone, kept in memory only for the comparison, and then discarded. They are never stored or uploaded, and we never use them to recognize or identify anyone. We only receive the pass or fail outcome and the score, to process refunds.

4. Who processes your data

Provider	Feature	What it receives
Google (Gemini API)	AI Studio image results (default); safety checks (moderation) of photos and anything you type	Your chosen photo, the style prompt, and text you type (for example an edit request)
OpenAI	Image model; safety checks (moderation) of anything you type	Your chosen photo, the prompt, the result, and text you type (for example an edit request)
Kie.ai and APIMart	Image processing: they route AI Studio requests to the image models (for example Google's Nano Banana and OpenAI's GPT Image)	Your chosen photo, the edit mask if any, and the style prompt
fal.ai	Fallback image model	Your chosen photo and the style prompt
Cloudflare (Workers AI, AI Gateway)	Fallback image model; routing, cost and latency logging	Metadata; your photo only when Workers AI handles a fallback
DeepSeek	Text assist only: tidies up text you type (looks, captions, edit requests)	Text and tags. Never images.

Which provider handles a given style can change as models improve; the app's AI consent screen lists every provider under "Privacy details", and this page is updated before a new provider receives photos. We do not train AI models on your photos, and we only use providers and routes whose terms do not allow training on the content we send. Providers may keep data for a short period (for example up to 30 days) for abuse monitoring.

5. Retention

Uploaded photos: deleted within 24 hours after the job completes.
AI results: deleted after 30 days, unless you back them up to your cloud library.
Results you save to your phone stay on your phone.
We do not use your photos or results to train AI models.

6. Safety and moderation

Photos and results pass automated safety checks provided by our AI providers and, where used, a separate moderation model. Requests that would produce prohibited content are blocked and any credits are refunded.
Styles are limited to light, color, scene and restoration. We do not offer styles that change body shape, age, ethnicity, or make clothing more revealing.
If a photo appears to show a child, AI Studio limits processing to color and lighting styles or declines it.
Every AI result has a Report button. Reports go to a human moderation queue. We act on valid reports by removing content, refunding, or restricting accounts.

7. Not allowed

Do not use AI Studio to:

create sexual or sexually suggestive images of anyone, or any content sexualizing or exploiting minors (we report child sexual abuse material to the relevant authorities);
edit someone else's photo without their consent;
create deepfakes or images intended to deceive, defame, harass or impersonate, including of public figures;
create hateful, violent or self-harm content, or infringe someone's copyright, trademark or likeness rights;
try to get around safety checks, credit limits or Identity Lock.

Breaking these rules can lead to removal of content, loss of credits and account termination (see our Terms).

8. Labeling and provenance

Some providers add provenance information to results (for example C2PA content credentials or a SynthID watermark). We try to preserve it when you save. In regions where AI-generated or AI-edited images must be labeled (including the EU from August 2, 2026), the app marks AI Studio results as AI-edited. Please don't present AI results as unedited photos in contexts where that would mislead people.

9. Credits and fairness

AI credits are included with Pro and are never sold separately. Each result's cost is shown before you confirm. Failed generations, safety blocks after credits were reserved, and results that fail Identity Lock are refunded automatically. Details are in our Terms and in the app's Credits screen.

10. Questions and reports

Email [email protected] or use Report on any result. For privacy requests, see our Privacy Policy.