coyotte

coyotte 

AI checkpoints creator, ComfyUI workflows designer

853subscribers

12posts

goals1
120 of 120 paid subscribers
Paid Subscribers
V9 in May. Z-Image alpha version might come out in May as well 👀
Cant wait to test it out. Did you have to retrain the captioning dataset from SDXL captions to the natural language captions of Z-Image?.
po vaj, No, I had just switched to the paid Gemini API completely. It's just miles better then any of the local alternatives and never refuses to caption images, no matter what I throw at it.
May is almost over. Can you tell if a V9 will come out in may?
No Name, alright, I think I'm happy with the version I was working on now. Will release it in a couple of days
coyotte, take your time, broheart
Comment removed
po vaj, I think you are confusing the new TOS with anima's licensing. I just read the new TOS dated May 28th, and there is nothing that claims that. Moreover, point 9.1 and 7 read "You retain any copyright and other proprietary rights that you may hold in the User Content that you Post to the Service..."
Apparently Ideogram 4 may be the new king of AI image generation. This guy has only trained an Ideogram LoRa for a few days, training run not even close to finished yet and already it is creating better quality NSFW than Flux Klein or Zimage that has taken months for them. Ideogram seems to converge incredibly quickly in training, it seems to respond lightning fast to lora training. His sample images are out of this world(vaginas still need more training time, but it will come), maybe in a month it will be indistinguishable from Met-Art style nudes. Unfortunately the base model will not be released as it is locked behind enterprise level commercial license. So training on the quantized model is going to be very hard as distilled models are bad for training/fine tuning. Interested to hear your take on it Coyotte. https://huggingface.co/RazzzHF/Realism_Engine_Ideogram_4/tree/main
po vaj, I see three big problems with ideogram 4:
1) Non-commerial licensing, meaning no way to raise the funds to finetune the model (as it's quite expensive to finetune. Big scale loras are also a challenge. Single concept loras are fine). And that also means the users aren't allowed to sell their outputs.
2) The license explicitly states they will take action against violations (finetuning it for nsfw IS considered a violation). Point "Use Restrictions" reads "However, we may also implement certain safety measures, content protections and other technological measures for the Model, including content filters and watermarking, and you agree that you will not circumvent, remove, alter, deactivate, degrade or thwart any such measures"
Meaning everyone who does so are shooting themselves in the foot, as in best-case scenario their loras/models with loras merged in will get removed; in the worst-case scenario, they could face legal action from the company.
po vaj, 3) We should be careful about praising models with restrictive licenses. If we embrace them, we risk setting a precedent where every new 'open-weight' release feels entitled to impose these kinds of limitations. We’re essentially voting for a future where 'open' doesn't actually mean open.
TLDR: I REALLY don't think it's worth the risk.
Ouch, I jumped on the band wagon too soon, sigh...guess its dead in the water then.
Subscription levels1

BASIC SUPPORT

$6.4 per month
An alternative option of donation
+ chat
Go up