Current best lora for fl2va is ref2va lora inference at 8step.
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
I'm still pulling off better results with drbaph's 8-step Rank 21 LoRA in comparison. More fluid movement and better action details throughout. Same iteration speed as well at 55s/it.
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
Which one are you referring to?
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
Which one are you referring to?
minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
Which one are you referring to?
minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
Hardly. The amount of prompt training I have had to do around that LoRA has been ridiculous. The current top two are silveroxides step600 and drbaph's 1.1 768p Rank 21 or 64.
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
Which one are you referring to?
minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
Hardly. The amount of prompt training I have had to do around that LoRA has been ridiculous. The current top two are silveroxides step600 and drbaph's 1.1 768p Rank 21 or 64.
I use the lora with 10eros or hybrid model, for me this lora retains facial consistency at high speed best.
I don't use a LoRA to maintain facial consistency. I attach an image and done. To each their own though.
No lora can actually rival this one, ref2va 4step lora 0.1 inference at 8step has the best prompt adherence, best facial consistency, best motion detection, best spacial detection, best audio prompt adherence, it can be used for both fl2va and ref2va workflow. Hope your team can finetune a real 8 step polished ref2va lora or universal lora.
Which one are you referring to?
minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
Hardly. The amount of prompt training I have had to do around that LoRA has been ridiculous. The current top two are silveroxides step600 and drbaph's 1.1 768p Rank 21 or 64.
Really? silveroxides lora generates overfitting ref2va video in my workflow and it even fails to understand proper scripts of different character, very poor audio quality, it also generates poor actions in combat scenes. the result is totally unusable compare to ref2va lora. I tried both loras applying on same base model.
Then it's your workflow, my man. I have a custom node with extensive prompt training and it obeys what I put into it.
Then it's your workflow, my man. I have a custom node with extensive prompt training and it obeys what I put into it.
could you kindly share it with me?
It's on my main repo page.