Subscription Updates: Usage Limits & Subscription Anlas Policy Adjustments

In the interest of full transparency, we want to address an important update directly.

With the launch of our latest image generation model NovelAI Diffusion V5 and the substantially larger architecture that a state-of-the-art model requires (2.5X the size of NovelAI Diffusion V4.5), we are facing significant compute constraints.

However, these constraints don’t come from ordinary users or a small percentage of power users alone. For the past year, we have poured countless hours into fighting abuse of our systems, and while that effort has kept things running, a small number of users continue to consume a hugely disproportionate share of our available compute and development time at everyone else’s expense.

Rather than raising prices across the board or introducing artificial generation delays, we’ve chosen measures that target the problem directly. To keep NovelAI sustainable while we continue to deliver bigger and better models and maintaining the high-quality experience you expect, we are introducing the following changes:

All Subscriptions

31 Days after the NovelAI Diffusion V5 release, we will change the way that the monthly allocation of Subscription Anlas behaves after you unsubscribe. After the announced date, any unused Subscription Anlas will be reset to zero the moment your subscription ends. If you cancel your subscription early, your Subscription Anlas will reset to zero on the date that the subscription would have originally ended.

Subscription Anlas have never stacked and your balance refilled to the full monthly amount rather than adding to it. So if you resubscribed after a break with Anlas left unused, they were being replaced by the fresh allocation without any chance to use them. Similarly, we were never able to restore a previous month’s Anlas when processing a refund.

Opus Subscriptions

For our Opus subscribers, we are introducing usage limits to our latest image generation model, NovelAI Diffusion V5 — all other models will remain unaffected with unlimited generations.

Here’s how it works
Think of your Opus usage limit like a rechargeable battery: every NovelAI Diffusion V5 generation (at normal resolutions and up to 28 steps) draws from your allowance, and that allowance recharges automatically, hour by hour, minute by minute, whether you’re generating or not!

A full recharge from completely empty takes about a week, and we’ve sized the capacity high enough that most Opus users will never see it run out.

Because the recharge never stops, a single heavy generation day, or a weekend marathon, won’t lock you out. Whatever you use starts coming back immediately, so you’re only ever hours away from generating again, never days. The recharge only pauses while your bar is completely full, which means regular creators naturally get the most out of the system: if you generate steadily, the recharge is always working for you.

You can keep an eye on all of this yourself

Your Usage Limit is visualized as a bar in your Goose Menu, showing your current charge, your recharge speed, and how long until you’re back at full.

If your usage limit ever runs out, you can keep generating images by using your available subscription Anlas or paid Anlas. As a reminder, Anlas generations are processed at the highest priority, so they remain the fastest way to generate on NovelAI.

Usage and Generation settings
Both resolution and step count affect how much of your usage limit a generation consumes. Both also affect the quality of the output. Reducing both will save more usage, but it may also affect quality more noticeably. Since cost scales with the pixel count of your chosen resolution, lowering resolution tends to be a more effective way to save usage than lowering step count.

**All previous Image Generation & Text Models
**All the image generation models before NovelAI Diffusion V5 and all of our text models remain unaffected: you can continue to generate unlimited generations at the default dimensions and settings, just like you are used to.

Thank you for your understanding, and for creating with us. As we continue to release newer models that require more resources, we need to establish parameters like these to keep NovelAI fast, fair, and reliable for everyone. It’s what lets us keep building bigger and better, and we can’t wait to see what you make with NovelAI Diffusion V5.

The Anlatan team