Anima in tensor art
Um why?
I know you guys need money and this model is really cool and good but makes need pro is insane look at civitai do they have the same thing no it's only It costs more money can we get the same like them I don't live in a country that allows buy anything from the Internet I am in Iraq
Greed kills the best models. Instead of striking normal deal with Tensor, author went this way, basically blocking people that cannot spend money and don't have good GPU to use his model.
See my comment here: https://huggingface.co/circlestone-labs/Anima/discussions/37#6a17cd154a0b7d3de1a39846
Instead of striking normal deal with Tensor
I do have a normal deal with TensorArt; they have a standard commercial license nearly identical to the one Civit has.
author went this way
I didn't choose to enable Pro-only, TA did that. I've emailed them to ask for clarification.
I didn't choose to enable Pro-only, TA did that. I've emailed them to ask for clarification.
I am not even shocked. That's very them too. I am very sorry you are getting the flame and apologize for being harsh.
More details in the other thread.
There is no direct requirement in the license to make the model Pro-only (or impose any other requirements). What there is, is a per-image license fee that necessarily increases the operational costs of hosting the model, and TensorArt has chosen to manage this cost by making it Pro-only. In principle I have no problems with this (or any other method a platform chooses to make hosting a model sustainable), but I'm still in discussions with them on the exact details of how it will look and what it means for end-users.
More details in the other thread.
There is no direct requirement in the license to make the model Pro-only (or impose any other requirements). What there is, is a per-image license fee that necessarily increases the operational costs of hosting the model, and TensorArt has chosen to manage this cost by making it Pro-only. In principle I have no problems with this (or any other method a platform chooses to make hosting a model sustainable), but I'm still in discussions with them on the exact details of how it will look and what it means for end-users.
Yea they don't like communicating very well at TA, and Civitai i am aware has a pro model but i have noted taht weirdly like flux: that odesnt' mean users get any kickback from it - I'm OK with this , i'm not complaining..
But holy crap i'm glad i kinda yeeted my ass off TA.
Someone's probably uplifted all my stuff and put it back - but not suprising considering i was consistently syncing my stuff and people still would reupload befor i could even put it up.
Hopefully you get this sorted.
Why are you blaming TA for trying to recoup the price of the license?
He got this sorted so well, that while it got off the Pro only, they increased gen price and made it permament credits only, making it basically paid only. I use TA because I cannot gen locally, using kaggle is also not viable. Before license kicked in, free users could gen easily and cheaply - his license basically is against free user on Tensor. Probably will see similar things in Pixai or Seaart if they ever get the license.
I guess I will post the updates in this thread. I previously said that the per-image license fee is $0.005. That is the default rate that I offer. In TensorArt's case, I negotiated with them and gave them a lower rate, to be more proportionally in line with typical end-users costs on their platform. I don't want to say the exact number as this is part of the deal between my company and TA, but it is significantly lower than $0.005 now.
As an example, TA's license fee for Anima is less than 25% of what BFL charges for Flux Klein 9b in their default self-service license (and Klein 9b is a heavier, more expensive model to self-host to begin with). If you calculate estimated self-hosting costs for rented GPU compute + the license fee, the combined cost for Anima is cheaper than almost every paid image generation model I can find, API-only or open-weights + license fee. There's like, maybe 3 models total on that list that might be cheaper than Anima.
At first, Anima cost 0.8 Credits per image on TA. As a reference, SDXL costs 1 Credit. Anima is 2.5x slower to run than SDXL due to its architecture, so by default it ought to cost more. At the beginning, TA was not only 1) in violation of the non-commercial license and not paying any fee, but also 2) heavily subsidizing Anima generations in relation to SDXL. Even ignoring license fees, this was an artificially low price charged to end-users that was never going to last, in my opinion.
TensorArt's commercial license is permissive, and they can choose to use the model and charge for it however they want. They pay only a per-image fee, nothing else is restricted or costs anything. They can allow whatever creator monetization programs they want. The license doesn't require the model to be gated, paywalled, or anything else. It doesn't charge for model training either.
I see a great number of people calling be greedy and unreasonable. I think the license fees I'm charging for Anima are very reasonable, and much lower than you would get from almost any other comparable model, based on the information I can find. I'm trying to build a sustainable business, and if I allow large platforms to use Anima for free, I will just go out of business and never train another model again. If you believe I'm being greedy and unreasonable, then please explain specifically what you think I should do differently.
sad but ok
I see a great number of people calling be greedy and unreasonable. I think the license fees I'm charging for Anima are very reasonable, and much lower than you would get from almost any other comparable model, based on the information I can find. I'm trying to build a sustainable business, and if I allow large platforms to use Anima for free, I will just go out of business and never train another model again. If you believe I'm being greedy and unreasonable, then please explain specifically what you think I should do differently.
While I understand the reasoning and think what you put on the table is fair, I want also to remind you that the success of a model is hugely linked to the distribution. I understand you wanting to get back your inversion, but trying to get it back too fast could be detrimental.
Even if you consider them fair or not, the thing is, it's more expensive than the competence. And the training is more limited. So far successful models on anime themes were not necessarily the best but the ones with more resources. Fewer people making LORAs for your model means less LORA, and less LORA means fewer people interested in the model.
So, I'm afraid the model could fall into obscurity, because even if it's good, if it has a limited distribution, it won't be a success.
No one can tell how the market will roll; maybe things will go fine, and I'm worried for nothing, but... you know... As I see things now, you could become the BETA of anime models.
So... you are in charge of the licenses, but people will tell. If people don't even bother making LORA for it because the trainers are very few instead of literally everyone, if people don't want to do gens with it because of the cost of the gens... well, not even all the amazing quality in the world is going to help.
Wish you the best luck, but take that into account. People may just settle with VHS
.... While that's a very hefty and clear statement, i think you're missing one large statement here:
People use TA, PixAI and Civitai all the time to train loras.
People are making Anima loras ON THE WEEKLY to the point i'm feeling like Illustrious is currently on the backburner.
Despite it's DIT nature, it IS Far better.
But we're also in a day and age that training a MODEL ITSELF is far more costly than it was in 2023, i'm not joking here when i say even ME using my training system to do Anima AND illsutriousXL --
I'm sitting at 50c an hour MINIMUM for A4500 and A5000/A6000 range GPU's.
And they're becoming more scarce as they're previous gen non consumer cards, and only just three weeks ago i was able to rent more 4090s at a reasonable cost.
And even AT the "SCARCE" costing, there's very little 4090s under 70c an hour at any time of my day, asuming last weekend it was "RUSH SWARM" and to find out nope - it was not.
So if TD Russel spent 50c MINIMUM an hour (try more like 3x that for a DIT arch full train) - and people are like "Y NO FREE" -- then why are y'all complaining like we're all socialist mobsters asking just a little kick back?
TA is banking it, Civitai is banking it.
If you want something "FREE" rent a GPU, use google colab..
Civitai doesnt even do IN workflow hi res fix anymore, I can't even explain what TA does because i literally deleted my account last time i had a big falling out in the AI community and almost deleted all of my Civitai account.
It IS VALID AS ALL GET out here that everyone wants something LOW cost if not mostly free cause y'know reasons -- but even a platform gotta keep up, so if they're GOING to get charged, they're going to put something under a PAID platform --
And honestly >_>
At this stage people should be less worried about TA and more the fact that half of Civitai isn't there anymore XD (or at least the OG staff have all rotated out lol)