BTCmix @btc_mix - Twitter Profile

@NEARProtocol @AskVenice Visible even with E2EE: Your IP, account, API key, auth session, model, provider route, request ID, billing/usage events, general metadata. Run a local model if you need full privacy. If you're happy with just E2EE, use imgnAI to avoid paying middle-man markups.

imgn_ai's tweet photo. @NEARProtocol @AskVenice Visible even with E2EE: Your IP, account, API key, auth session, model, provider route, request ID, billing/usage events, general metadata.

Run a local model if you need full privacy.

If you're happy with just E2EE, use imgnAI to avoid paying middle-man markups. https://t.co/p1n4aSCDKe

0

22

9

1

985

btc_mix retweeted

imgnAI

@imgn_ai

26 days ago

Tomorrow, we'll be making some changes to our Free Tier offerings on imgnAI. Over the past 3~ years, we've always maintained an aggressively generous Free Tier, allowing for unlimited gens (at HD/QHD quality levels) for all users on our in-house models. This was done with the idea that these free users would upgrade to our Premium tiers - and unfortunately, the past 3 years of data don't reflect this idea. Users who upgrade to our Premium tiers tend to do so either before generating at all, or after a small few test gens. When they go beyond that, the chances of them converting to a paid tier are effectively zero. The flip-side was the expectation that users running with free tiers would "self-advertise", bringing in new users who may pay. This also didn't turn out to be the case; with over 99.9% of gens going completely unpublished. Our endpoints have always been popular; generally we'd max out at around 1 million images generated per week. This is around our max capacity; any greater demand, and queue times appear. For context, this is around the "compute equivalent" of 100 Billion tokens on an 8B LLM. Even with priority levels in place for Premium users, without literally throwing gens out of a GPU mid-render, these queues got noticeable. Pink Image for example renders in <15s; but yesterday it exceeded 45s due to free users in the queue. While we're going to maintain the ability for users to test run our models as intended, we're going to impose daily limits of around 20 generations per user on unpaid/free tiers. These limits will be drastically raised for paid tier users who want to iterate on some free gens; with 200 free gens per day for silver tier users, up to 5000 free gens per day for our pro tier users. Additionally, a mandatory queue will be applied on any free tier gens which is based on the overall amount of site traffic that we're handling. Currently, this wait is set to around 3 minutes for a batch of 4 image gens. Note that these numbers are subject to variance based on system load. We want to ensure that queue times are non-existent for our paying customers, along with ensuring that our GPUs are being put to the best use possible in improving our services when there aren't paid requests to handle; such as training new models, or improving on existing ones. We've a lot of new products that we'll be rolling out over the next few weeks and months, which we expect to continue increasing the load on our hardware - and our goal is to ensure that the value gained from that can be directly returned to our paying users and token holders, rather than investing in hardware which serve more towards a free user subsidy than a value add. We're extremely grateful to all of our supporters who have to date, made these free services possible to begin with. We look forward to repaying the favor with our upcoming new releases.