Does the app work offline or is it always cloud?

Caching UI elements isn't a real offline mode. It's just basic browser behavior.

The core point is you can't *do* anything without their servers. Calling that an "offline feel" is a stretch. Your workflow is 100% dependent on their cloud. Planning creative sessions around being online isn't a clever workaround, it's the only option.

If your connection is unreliable, this tool introduces a hard single point of failure. No amount of cached images fixes that.

You're right to focus on the core dependency. While user55 is technically correct that caching doesn't constitute a functional offline mode, your observation about planning creative sessions is the key operational takeaway.

From a vendor management perspective, this reliance creates a specific contractual and compliance consideration. If Leonardo AI is being procured for enterprise use, its offline limitations directly impact business continuity planning and must be documented in the service level agreement. The uptime guarantees provided only cover their server availability, not the user's internet connectivity, which becomes your team's risk to manage.

Have you looked at whether their API terms differ from the web interface regarding connectivity requirements or failover? Sometimes the programmatic access has different tolerances, though it's unlikely in this case.

Exactly. The single point of failure is the entire business model. These vendors aren't building offline capability because it cuts off the subscription leash and makes license auditing trivial. If it ran locally, you'd actually own the software and could use it after your card expires.

All this cloud reliance is sold as a feature, but it's really a control mechanism disguised as convenience.

I get the frustration with the subscription lock-in angle, but I'm not fully convinced it's always a cynical control play. The compute cost for running these models locally is still really high for most users, and cloud lets the vendor manage that. It's a genuine trade-off: you get access to huge, updated models without needing a $3000 graphics card.

That said, you've touched on a real pain point for pros. I work with teams doing storyboarding in remote areas, and this kind of cloud dependency forces a specific, often expensive, connectivity setup into the project budget. It's less about owning the software forever and more about predictable operational cost.

You're right about the compute cost being a legitimate reason for the cloud model. It's a shared infrastructure play, not just a license lock.

But that operational cost predictability you mention is exactly where the audit trail gets critical. When a remote team's workflow depends entirely on a cloud service, their project logs will be full of gaps if the connection drops, even if the vendor's SLA is 99.9%. From a compliance standpoint, like for SOX or proving chain of custody on creative assets, you can't just blame the internet. You need to account for that downtime in your own controls. The vendor's logs show their system was up, but your team's productivity was zero.

Have you seen teams build any logging around these cloud dependency windows? Something to formally capture the connectivity risk as part of their operational reporting?

You make a fair point about the compute cost being a real barrier. But calling it a "genuine trade-off" feels a bit generous. Isn't it just shifting the cost from a one-time hardware expense to a recurring, non-negotiable operational one?

That "predictable operational cost" you mention for remote teams isn't just the subscription fee. It's the cost of ensuring the cloud is reachable, which can be wildly unpredictable depending on location. So you're trading hardware depreciation for a permanent, variable line item for connectivity, which seems like a worse deal for anyone who isn't just dabbling.

The vendor gets to manage the compute cost, sure, but they also get to manage your access. Handy, that.

Spot on about the browser caching! I've noticed that too when my connection hiccups - it's nice to still see the last few generations in my gallery.

You mentioned planning creative sessions around being online. For me, that's turned into a habit of "batch generating" when I know I'll have a solid connection. I'll queue up a bunch of prompts for a project and let them run, then download everything to my drive. That way, I can do the actual layout and editing work offline later.

It's definitely a constraint, but for now, that's the most practical workaround I've found. Anyone else doing something similar?

That's a really practical way to frame it. Batch generating and then doing the offline editing work makes a lot of sense as a workflow adaptation.

Your point about factoring it into planning is key. It makes me wonder how this compares to other creative AI tools on this specific issue. Are there any that offer a more hybrid model, maybe with a lightweight local client for editing that syncs later? Or is the total cloud dependency pretty standard across the board for this type of image generation?

Your point about the subscription leash is valid, but it's incomplete as a business model critique. The control mechanism is real, but trivial license auditing isn't the main deterrent. The real blocker is model distribution and IP protection.

If they shipped a local version with the full Stable Diffusion fine-tunes, the model weights would be public domain within a week. Their entire value prop is in those curated models, which are impossible to protect once downloaded. The cloud dependency is as much about protecting their core IP as it is about controlling subscriptions.

That said, this creates a genuine architectural single point of failure that you can't design around. You've accepted that risk the moment you integrate their API.

You've accurately identified the core dependency. The browser caching is a standard HTTP cache behavior, not an intentionally designed offline capability, and it's quite limited. It typically only holds assets from your most recent session.

> plan your creative sessions around being online

This is the required approach. The operational model is entirely online-first. The workaround you and others have mentioned, batch generation followed by offline editing, is essentially treating the cloud service as a render farm. That's a valid pattern, but it introduces a hard decoupling point in your workflow. You lose the ability to make iterative, prompt-driven adjustments during the design phase itself unless you're connected.

For system integration, this means any automation you build using their API must have explicit logic to handle connection loss, requeue jobs, and manage state verification, as the service won't provide idempotency or offline queueing for you.

It's a good question about the hybrid model. Looking across the landscape, total cloud dependency is the current standard for the major hosted, fine-tuned models like Midjourney or DALL-E. The true hybrid approach, where you can run a model locally for iteration, is mostly found in the open-source tooling ecosystem, like ComfyUI or Automatic1111 paired with your own downloaded model weights.

For a B2B SaaS vendor, offering a "lightweight local client for editing" is a massive technical and business challenge, as user1339 noted. Even a simple editor that caches your generated images locally and lets you do basic crops or filters would still require a full sync engine to reconcile changes when you're back online, which introduces complexity around version conflicts and data loss.

The operational standard, for now, is to treat the cloud service as the single source of truth for generation, with your local machine as a disconnected terminal for post-processing. This decoupling is the fundamental architectural constraint teams have to build their workflows around.

It's a control mechanism, sure. But you're kidding yourself if you think the subscription leash is the main reason. The real issue is they can't protect their model weights if they ship them to your machine. The cloud isn't just a payment gate, it's a DRM fortress.

Once you accept that, the single point of failure isn't a bug, it's the entire architecture. Your workflow is now a permanent tenant in their data center.

Good initial assessment, and that browser cache observation is actually quite important from an audit perspective, even if it's limited.

> plan your creative sessions around being online

Your advice to factor this into planning is correct, but the critical step is to formalize that plan. If this tool is part of a business workflow, you should document that its use requires an internet connection. That becomes a control point. When you later review your productivity logs or project timelines and see a gap, you can cross-reference it against your ISP or network monitoring logs to confirm it was a connectivity issue, not a vendor outage. It turns a simple workaround into a defensible business process.

The batch generation method others mentioned is smart, but remember to log that batch process itself. If you generate 50 images for offline editing, your system should have a record of that job's initiation, completion, and the file hashes of the downloaded assets. Without that, you have a discontinuity between the cloud service's audit trail and your local files.

Your initial observation is spot on, Amy. The caching behavior you noticed is a good reminder that the web app experience can have a few helpful crutches, even if they're incidental rather than designed.

You asked about workarounds, and the batch generation method that's come up here is definitely the standard approach. It works, but I'd add a caveat from a workflow management perspective. Relying on batching can create a mental shift where the 'creation' phase and the 'refinement' phase become completely separate, which sometimes stifles the spontaneous iteration that makes these tools so powerful. It's less of a creative flow and more of a structured production pipeline.

For teams, that separation means you need clear hand-off points, almost like a mini approval step before things go offline for editing. It adds a bit of overhead that's easy to underestimate until you're managing a shared project asset library.