I want even faster
感谢你的项目,原作者最新的850lora我也用了,你的执行起来会更快一些,然后我还想进一步提升速度,我设置的分辨率0.4, 10s的视频, 执行时间能否再加速一些, 我现在比4060的笔记本电脑还慢, 已经开了虚拟内存, 并使用了这个项目的lora(我改了名字,下载的加速lora太多了不好区分)
Thank you for your project. I've also used the original author's latest 850 LoRA, but yours runs faster. I would also like to further improve the speed.I set the resolution to 0.4 for a 10-second video. Can the execution time be further accelerated? Right now it's even slower than my laptop with a 4060. I've already enabled virtual memory and am using the LoRA from this project (I renamed it because there were too many downloaded acceleration LoRAs to tell apart).
thx!
Use this workflow https://huggingface.co/Abiray/MiniMax-H3-Turbo-Lora-Pruned-ComfyUI/blob/main/minimax_h3_t2v_turbo.json settings, and if can update the MiniMax H3 sampler version https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3/releases/tag/v0.1.8 with new Spectrum settings
New default settings
degree = 1
warmup_steps = 1
bootstrap_first_forecast = true
tail_actual_steps = 1
hi, Abiray. I have a question
do you have any plan to convert these turbo loras for base models - not for pruned only, because I want use step 600 turbo lora with non-pruned model without any Larryvrh nodes
This LoRA model version requires no additional nodes and works with both pruned and non-pruned models. Simply use the workflow available in my repository.