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

anything Wan 2.2 can do, Minimax h3 can do better. 10eros hybrid model can handle i2v, t2v, and ref2va without swapping models. Look for the int8 version and you should be good.
 
Has anyone managed to use these ai assistants to make image or video generation more optimum ? Like can it create optimised workflows ? like people are making entire games with the help of ai assistants with 0 coding knowledge.
 
ok so now I managed to shrink generation time, but when i give comfyui a video + photo references of a person it just remakes the video with the subject from the video, without actually swapping for the person in the pictures i gave, any suggestions? I tried asking gpt or grok but they were of no help
 
Look at civitai for workflows that specify low vram, keep your initial render small probably ~640x480 and you should be fine for shorter clips.

You may also want to check out LTX 2.3 (not sure about 2.5), its still a decent model and way lighter on resources than minimax.
 
Yo!... I have the same NV 3070 with 8 vram as you I guess (or similar), I have 32 ram thou.
I found it nearly impossible to make it to work with a resonable timelimit of generate a 5,3 sec vid, took forever, tried several ComfYI loadouts with same crap, so I switched to the WanGP (BeebBeep) interface instead wich is kind of an an easier/web-buid interface (this was just before Turbo/ 4 steps was released thou so I didnt tried it out within the ComfYI).... to summarize...

ComfYI is superior for use as a scaled shellbased interface (I use it alot in other AI,s) but if your hardware isnt the top of the line you need drop down a bit...

WanGP gives you an easier kind of way to generate vids (for my card NV3070) up to 10 sec without any problem (720p, about 17-25 min with Turbo, 4-steps, with 15 sec vids about 45-70 minutes), 1080 gets trickier and muuuuuuch longer generations so I dont even try.

You can use the most popular loras due to the regular updates as of (no Civit loras mentioned thou, but you can download and use them as well):


Just choose as you like.

Conclution, if your stuck within the ComfyUI, give WanGP a try, works fine for me and Iam happy considering my hardware ;)

/AK5
 
Preferable is a Nvidia card, NOT AMD, but Iam not sure if that has changed lately thou. Nvidia card is the "golden solution" to manage a "flawless" generation of happiness in this AI world.
 
This. I literally can't do shit until a fix is put in place. AMD specific portable Comfy can't launch python if you don't have a D drive. For some reason it's hardcoded so even dicking around isn't an easy fix. Tried some stuff and gave up. Just praying for an update/fix.
 
Anyone know a good LLM for creating prompts? i'm using claude but it keeps stopping my edits if it thinks i'm using a real person as a reference.
 
You must log in or register to reply here.
Back
Top