AILimited operational AI · NO AI conversion · NO training on uploads.

AI Policy

Last updated: May 2026

The AI question for an image-conversion site has two parts: does the Site use AI in its conversion pipeline? And does the Site use uploads to train AI? Headline: no AI in the conversion pipeline (we use traditional libraries: ImageMagick, libvips, libheif, libavif, libraw); we do not use uploads as training data; AI use on Site content is opted out.

The five positions

Limited operational AI use is permitted for editorial drafting (with substantive human review), research, and routine tasks where AI helps acceleration without affecting quality.
Image conversion does NOT use AI. Our conversion engine is traditional libraries. We do not run "AI upscale", "AI enhance", "AI denoise", or generative-restoration on user uploads.
User uploads are NEVER used to train, fine-tune, or evaluate any AI model. Period.
Editorial content is human-written. The trust pages, format-catalog descriptions, and tool copy are written or substantively edited by the operator, not auto-generated and shipped.
Site content is opted out of AI training by third-party AI builders. The opt-out is asserted contractually and implemented technically.

1. Why no AI in the conversion pipeline

"AI image processing" is a real and growing category — for upscaling, denoise, super-resolution, content-aware cropping, generative restoration. We don't do any of it. Reasons:

It's a different scope. iloveimg.online is a format converter, not an image enhancer. Adding AI processing would change what the Site is.
The deterministic-format-conversion guarantee. When you ask us to convert PNG to JPEG, you want the same image in JPEG — not "the same image, but our AI thinks it should be sharper". Traditional libraries are deterministic; AI is not, in the same way.
Privacy implications. Many AI inference services involve sending the image content to a third-party API (OpenAI, Google, Anthropic, etc.). That's incompatible with the data-handling commitments on Data Handling: server-side conversion in our control, deleted within an hour. Sending uploads to AI APIs creates a different trust party.
Quality predictability. Library-based conversion has well-understood failure modes (lossy quantisation, format-specific artefacts). AI-based processing has less-predictable failures (hallucinated detail, content misinterpretation).
Cost. Running AI inference for every conversion would change the economics. We'd have to charge or accept much heavier ad density.

If you specifically want AI upscale, denoise, or generative restoration, dedicated tools exist: Topaz Photo AI, Adobe's Enhance, Let's Enhance, BigJPG. We're not in that business.

2. What "limited operational AI" means

Permitted operational uses:

Drafting trust-page copy (this set of pages). AI is used as a drafting tool; output is substantively edited by the operator before publication. No "AI dump" copy.
Research on format specifications, library options, regulatory frameworks.
Code review of non-conversion infrastructure. Laravel customisations, deployment scripts.
Translation as a working tool for non-English source material.
Spell-check / grammar tools with integrated AI features.

3. Where AI is not used

The conversion engine. Traditional libraries only.
Reader-correspondence responses. Substantive email is answered by the operator.
Format-catalog descriptions presented as canonical.
Recommendations to use specific competing tools. When we recommend ImageMagick, GIMP, or another tool, the recommendation is editorial.
Image-content moderation. No AI moderation pass on uploads. This is a privacy choice (and per Responsible Use, content responsibility rests with the uploader).

4. AI training on user uploads — absolutely not

This is worth stating clearly: uploads to iloveimg.online are not used to train any AI model. Specifically:

Uploads are not sent to AI training services.
Uploads are not retained beyond the one-hour conversion window for any purpose, including training.
Uploads are not aggregated into training datasets.
Uploads are not used as input to any "let me see what kinds of images people convert" data-science effort.
This commitment applies regardless of any future change in AI infrastructure pricing or commercial pressure.

5. AI training opt-out for Site editorial content

Use of the Site's editorial content (this 22-page set; format-catalog descriptions; tool copy) for the training, fine-tuning, evaluation, or other development of AI / ML / LLM systems is expressly prohibited.

Legal basis

EU Directive 2019/790, Article 4(3) — the text-and-data-mining exception is conditioned on the rightsholder not having reserved rights in an appropriate machine-readable manner. This Site reserves those rights. Italian transposition: D.Lgs. 177/2021.
U.S. Copyright Act, 17 U.S.C. § 106.
Italian copyright law L. 633/1941 as amended.
Equivalent rights in other jurisdictions.

Technical implementation

The Site's robots.txt blocks known AI crawlers. The list (non-exhaustive):

GPTBot, ChatGPT-User (OpenAI)
ClaudeBot, anthropic-ai (Anthropic)
CCBot (Common Crawl, used in many training pipelines)
Google-Extended (Google AI training; distinct from Googlebot for search)
Bytespider (ByteDance)
FacebookBot / meta-externalagent (Meta AI)
Amazonbot, PerplexityBot, Diffbot, Applebot-Extended

Important nuance: search vs. training

The opt-out is for training use.
Standard search-engine indexing (Googlebot for Search, Bingbot, etc.) is permitted.
The distinction matters because some operators (like Google) operate separate crawlers for search (Googlebot) vs. AI training (Google-Extended).

6. Why this position matters

Trust. Users uploading images to a converter need to trust the operator. "Your uploads are not training data" is the floor for that trust.
Format-fidelity reliability. Deterministic library conversion produces predictable results.
Reader expectations. A converter is expected to convert, not enhance.
AdSense compliance. Original, human-written editorial content supports site quality.
Regulatory alignment. EU AI Act and emerging frameworks favour transparency about AI use; this page provides it.

7. Updates

This Policy is reviewed periodically and updated as the underlying technology and regulation evolve.

Related pages: Copyright Notice · Data Handling · Format Catalog · About Us · Our Approach