text2vid img2vid H3 Minimax - Fully uncensored local AI video generator released today

Only issue is no pussy knowledge so it makes a slit/hole, it's not that bad but it's not right, adding a ref image for it helps but sometimes gives weird results.
 
using minimax_h3_fl2va_pruned_int8_convrot and qwen3vl_32b_minimax_h3_int8_convrot
 
Been using ComfyUI to make stuff using Reference to Video but there is some jank. I'm only using the default template so is their a recommendation for settings? I asked Venice and it fixed some blurring and artifacts but maybe it's my prompts.
Still, anyone have settings that are absolutely perfect?
 
I tried on Maestro on my PC with a 4090 and 64GB of RAM. It was just the base model. I tried 480p 10 seconds. It used all my RAM and like 21GB of VRAM and what it put out was nightmare fuel. I'm gonna try it on Comfy.
 
Try it with the models commented here:
Those are optimized and shouldn't take as long
 
i updated comfy and still only get API templates besides the minimax, but wanna try grok, does anyone know what else to do ?
 
I think my model is somehow moderated, cause it adds underwear to the woman all the time even if I prompt to not do it. I am in comfyui through runpod with the template One click comfyui minimax h3. The results are awesome tho, close to grok.
 
Are you using the workflow that comes with the template on runpod? Or did you select the one that ComfyUI made and it's under the Templates tab inside ComfyUI?
 
How good are the results compared to something like SuperGrok video generation?
 
They are good. But if U use local version U should spend some time to find right promt
 
You can probably get pretty close visually, at least with references, but it's a lot more finicky currently. The thing about Grok was that it usually just did all the things without you even asking because it just wants to make porn.
 
I am using the workflow with the template on runpod yes, but aparently adding a new encode more uncensored called "heretic" fixed de problem, and now the results are better. Now I see that the model does not know the anatomy of the penis and vagina...I guess that will be fixed with LoRas in the future.
Another "problem" that I saw is that I cant make them having sex fast, is like they are in slow motion or something, a little bit robotic. I am currently using only i2v.
Edit: it seems that the longer the duration of the video is, the worse the result will be. 5 secs are good but more than that not so much.
 
the sound needs A LOT of work, but gotta say this is a pretty good start. eventually people will come in with models/LORAs to improve upon this. there is already a turbo LORA to significantly increase processing time. the gooning potential is high here
 
New to comfyui how would i go about adding a lora to the text to video template to get a specific look or video output?
 
Download a Lora u want, put it in the Lora folder within Comfy map, add Lora node (cant remember the name atm) to the workflow, choose the specific Lora within that addon, dont forget the "trigger words" (look for this from the uploader on the site u downloded the Lora from) in the promt due to the fact they help out for better results.
 
I havent really touched anything besides adding a lora (turbo) and connecting in the default "use image size" provided. But its pretty good for being 3-4 days old.
Like this will be awesome in a few weeks. Currently, as others have written, the sound is a bit janky and the prompting is so-so, but again, 3-4 days old. Its nice =)
 
Is it possible to skip generating the audio or reduce the FPS without it just slowing down and extending the video? Im using the template from ComfyUI and I just want to speed it up a bit
 
You must log in or register to reply here.
Back
Top