Skip to content
Notifications
Clear all

Does the 'private generation' setting actually mean private?

3 Posts
3 Users
0 Reactions
50 Views
(@charlie99)
Reputable Member
Joined: 4 months ago
Posts: 310
Topic starter   [#28439]

Hey everyone! I've been deep-dive testing Leonardo AI for a few weeks now, primarily using it to generate conceptual visuals and mockups for data pipeline architectures (think: flow diagrams, stylized database icons, etc.). It's been a fantastic tool for that.

Naturally, some of the concepts I'm generating are for proprietary systems my team is designing. I saw the "Private Generation" toggle in the dashboard and, like any good data engineer thinking about access controls, I immediately switched it on. But it got me thinking... what does "private" actually mean in this context? From a data integration perspective, I'm used to terms like "private" having very specific technical and legal definitions, especially when dealing with cloud services.

So, I'm hoping we can pool our knowledge here. I have a few specific questions, and I'd love to hear about your experiences or if anyone has dug into the Terms of Service:

* **Does "Private Generation" simply mean the image isn't displayed on the public Leonardo feed/galaxy, or does it also imply something about data processing?**
* **Are the prompts and image metadata associated with a private generation also kept out of any training data or improvement algorithms?** In the ETL world, we'd call this "data isolation" or "tenant segregation."
* **What about the underlying image data itself?** Is it stored on a separate, more secure infrastructure, or is it just a flag in a database row that hides it from the UI?
* Has anyone encountered any transparency reports or data processing agreements from Leonardo that clarify this?

I tried looking for an API endpoint hint or something in the network calls, but no luck yet. The reason I'm so curious is that in my world, if a data pipeline isn't *truly* private, you need to anonymize or tokenize the input data before sending it to an external service. If I'm generating an image based on a confidential system design, the prompt itself could be sensitive.

For example, a prompt like: `"a detailed diagram of a real-time fraud detection pipeline using Apache Kafka and a proprietary model 'X-247', neon cyberpunk style"` contains potentially sensitive project names and architecture intent.

If anyone from Leonardo is lurking, some clear documentation on the data lifecycle for private generations would be incredibly helpful! It would influence how and when I use the platform for work-related projects.

Data nerd out


Data nerd out


   
Quote
(@alexg)
Honorable Member
Joined: 4 months ago
Posts: 564
 

You've hit on the absolutely critical distinction that most vendors hope you won't think about. The public feed exclusion is the easy part. The data processing and retention angle is where the real contractual and technical substance lies.

In my experience with cloud service SLAs, "private" in a feature toggle context almost never extends to the training pipeline unless explicitly stated. You need to scrutinize their data processing addendum, not just the main Terms of Service. Look for clauses on "service data," "content for improvement," and opt-out mechanisms for AI training. Many platforms consider your prompt and the generated image as separate data entities, with different retention policies for each.

Your second bullet point is the real question. It's common for a platform to keep your *image* private from other users while still retaining the right to use the *prompt text* and associated metadata for model fine-tuning or analytics. I'd recommend opening a support ticket and asking for a specific, written clarification on the data lifecycle for private generations. Get them to define the terms.



   
ReplyQuote
(@henryf)
Reputable Member
Joined: 4 months ago
Posts: 291
 

Exactly. The prompt/image split is key. I've seen similar in CI/CD logging where a vendor calls build logs "private" but still scrapes metadata for product analytics.

Without a written opt-out clause for training, assume it's fair game. Even then, you're trusting their pipeline cleanup.



   
ReplyQuote