How long does the approval process take?Can it be made public before the release of MiniMax H3.1?
Let me think about the timeline. MiniMax H3.1 should be released within the next two months, right? Based on what the MiniMax CEO said previously, it should feature a larger parameter count and integrate the language-understanding capabilities of MiniMax's own language models to achieve better unified instruction following.
So, when is this currently unreleased project expected to launch? Are there any users who have already tested it? Is this derivative model actually worth waiting for? Are there any real-world demonstrations of its performance?
Let me think about the timeline. MiniMax H3.1 should be released within the next two months, right? Based on what the MiniMax CEO said previously, it should feature a larger parameter count and integrate the language-understanding capabilities of MiniMax's own language models to achieve better unified instruction following.
So, when is this currently unreleased project expected to launch? Are there any users who have already tested it? Is this derivative model actually worth waiting for? Are there any real-world demonstrations of its performance?
It is in early access to those who donated $100 or more for the model to be trained. Also it is a H3 model tuned specifically on porn to make porn. So it is 100% community funded and not funded by any known investors. I would expect at minimum at least another week before that early access period for donors is done. Takes a lot of testing and R&D for any good full fine tune or even good lora to be made. And to my knowledge Sulphur 3 will require much different prompting and sampling methods than normal minimax h3 since it changed the model so much during training. H3.1 could also be an API only model knowing how the video model space works and the pattern of diffusion model developers. I mean the flux video model is still API only several months later after being released and may never actually go open weights. So its possible minimax released H3 for free to get their name out there more in the video model space to be able to sell API access to future H3 models.
In my case, i saw an advertisement for Sulphur 3. That is why i came here. Why run advertisements if the lora isn't available?
In my case, i saw an advertisement for Sulphur 3. That is why i came here. Why run advertisements if the lora isn't available?
If your talking about the bot post from the HuggingModels twitter uh yea thats very hallucinated. So you are not blind sided when sulphur is out just know it is a NSFW finetune of Minimax H3.
Not sure if this is the case but that ad might have been from a unofficial source thats not even related to the Sulphur 3 team trying to build hype so when it is released they can sell API access to it on their platforms.
So this is just a hardcore sex action porn finetune of minimax h3 or are is there other nsfw concepts trained in like morphing, shapeshifting and body transformations?
So this is just a hardcore sex action porn finetune of minimax h3 or are is there other nsfw concepts trained in like morphing, shapeshifting and body transformations?
It is a gooner finetune of h3 also im not sure if morphing, shapeshifting and body transformations classify as NSFW. Sulphur 3 is for booba and vagoo.
please accept me
please accept me
Likely going to need to wait a few weeks
H3.1 is coming out soon (according to minimax investors) people for sure will flock to new model so there's no point to wait with opening Sulphur-3. Just please open the model.
H3.1 is coming out soon (according to minimax investors) people for sure will flock to new model so there's no point to wait with opening Sulphur-3. Just please open the model.
Well Sulphur-3 is a gooning model and H3.1 is a new H3 which is great and all but its likely not a replacement for a NSFW finetune.
Well, this fine-tune is very specialized for a specific type of action scenes, camera work, etc. The base model is already well prepared for this kind of video generation. What can H3.1 bring to the table? More dynamic action, better fight scenes, longer videos? All of that is more or less already covered by LoRAs, nodes, and other stuff. So I highly doubt this fine-tune will lose its value if and when H3.1 comes out.
Not to mention how much they might dumb down or nerf the new version when it comes to the kind of content we're talking about.
Well, this fine-tune is very specialized for a specific type of action scenes, camera work, etc. The base model is already well prepared for this kind of video generation. What can H3.1 bring to the table? More dynamic action, better fight scenes, longer videos? All of that is more or less already covered by LoRAs, nodes, and other stuff. So I highly doubt this fine-tune will lose its value if and when H3.1 comes out.
Not to mention how much they might dumb down or nerf the new version when it comes to the kind of content we're talking about.
Better base model menas better physics and an better world knowledge which also translates into better NSFW content. Plus we might even get a raw model. In that case, what’s the point of using Sulphur-3 when there will be LoRAs for H3.1 that can execute scenes much better?
Well, this fine-tune is very specialized for a specific type of action scenes, camera work, etc. The base model is already well prepared for this kind of video generation. What can H3.1 bring to the table? More dynamic action, better fight scenes, longer videos? All of that is more or less already covered by LoRAs, nodes, and other stuff. So I highly doubt this fine-tune will lose its value if and when H3.1 comes out.
Not to mention how much they might dumb down or nerf the new version when it comes to the kind of content we're talking about.Better base model menas better physics and an better world knowledge which also translates into better NSFW content. Plus we might even get a raw model. In that case, what’s the point of using Sulphur-3 when there will be LoRAs for H3.1 that can execute scenes much better?
No one knows if they are going to aggressively censor H3.1 and I have only seen mentions of H3.1 in MiniMax’s August 26 earnings call.
Well, this fine-tune is very specialized for a specific type of action scenes, camera work, etc. The base model is already well prepared for this kind of video generation. What can H3.1 bring to the table? More dynamic action, better fight scenes, longer videos? All of that is more or less already covered by LoRAs, nodes, and other stuff. So I highly doubt this fine-tune will lose its value if and when H3.1 comes out.
Not to mention how much they might dumb down or nerf the new version when it comes to the kind of content we're talking about.Better base model menas better physics and an better world knowledge which also translates into better NSFW content. Plus we might even get a raw model. In that case, what’s the point of using Sulphur-3 when there will be LoRAs for H3.1 that can execute scenes much better?
No one knows if they are going to aggressively censor H3.1 and I have only seen mentions of H3.1 in MiniMax’s August 26 earnings call.
Yes I agree we won’t know many of the technical details until the model weights are released. However Minimax has no incentive to censor its models in the way Black Forest Labs does. Many companies even major ones like OpenAI don’t necessarily censor NSFW training data by default instead they employ intelligent filtering systems. Models like GPT-Image, Nano Banana etc. understand what NSFW content is, but the filtering system simply blocks the generation of such images. Any sensible company uses this approach rather than stripping out "inappropriate" data from the training dataset. As for the information itself these aren't just leaks or rumors they are actual data points from Minimax's investors. If you dig deeper into the Chinese internet you can find much more information. I’m not entirely certain but the news about significant improvements in physics has been circulating for a few months now.