Testings Out Minimax H3 LORAs and extending single shots. This is a 40 sec seamless shot, still getting the hang of this but this is dope !

Very cool. Lots of continuity issues (I am sure you're aware) her saying hello before picking up the phone could easily be removed but walking from the phone to another phone took me out of it. Does this model take references? I often send an overhead view of the location as reference and it really helps lock it in. Truly creepy! Love it!

It’s a true horror scene. It was the ghost that moved the TV, not the AI!!😅

Yes, he was altering everything that is why she freaked out your right ! THATS THE STORY !

If your video generator takes references create a location reference. Ask for a 3/4 view of the room you want (describe in detail), it helps a ton with continuity issues. Your aesthetic and monster are really freaking awesome! Just lock your location asset and you’ll be golden.

This is awesome Tony. This on the sparks? Please share what you’re using for the extension. Is it civitai and the 80s horror Lora?

this is on the 3090 but the same thing can be done on the sparks as far as the LORA ya

Yeah wiring up now. How fast you generating on 3090? Faster then on the spark Right?

hey man what version of this do you recommend on a 2 spark setup to get the maximum best fastest quality

one shot? workflow @Tech2Wild ?

this is cool, what workflow did you use to do seamless extension?

A seamless single shot that long is wild. How are you extending it, feeding the last frame into the next generation or something the LoRA handles?

You scared me before going to bed. Amazing video. It is a movie right here.

How?

40 秒接得上很爽，但 H3 长镜头的分水岭不是首帧，是中后段身份、手部和光线能不能守住。LoRA 让时长变长不等于叙事变稳；连贯性不过关，长镜头只是把穿帮放大。

It’s easy all you do is just make a prompt

That's impressive. Dig it.

Assuming you had it stitch multiple shots for this?

workflow plz 🙏

This is cool which workflow

Which Loras are these and what time it took to generate? Happy to know the workflow

Why is she sitting on the side of the tv, is H3 retarded?

You gonna cook up a recipe to the public?

Jeez how long did this take?!

what lora is this?

The biggest problem with ai videos now. The spatial awareness. It does not follow where the phone was so you need to guide it. And it takes a lot of time.

This is great. I'm currently renting GPU on Runpod and building out my Comfy UI/Minimax set up there. Is this a LORA you built?

On sparks??