settings?

#2
by Rkss - opened

which settings should we use
Sampler, scheduler, lora strength, model shift if any ?
can you make one for ref2va too?

i'm interested too

dareties works great for r2v also.
use the settings, 0.8 strength, shift 8/3, 6-8 steps, euler simple, or LCM simple.

er_sde ODE with eta 0 / power_shift seems interesting too

guys i am getting this error:

ERROR lora diffusion_model.blocks.15.adaln_proj.linear.weight shape '[96768, 8]' is invalid for input of size 260112384

and it keeps going, using int8 pruned official..

@necroryona

no expert but followed PlagueKind workflow and it fixed my error, if it the correct way? not sure but it did remove the error:

  1. create folder: h3_adaln in "ComfyUI/models" folder
  2. https://huggingface.co/deAPI-ai/minimax-h3-33b-int8/resolve/main/loras/h3_silu_temb_grid.safetensors put this file in h3_adaln folder
  3. image

put this node after your custom lora loader node(if using)

I have that node in my pack

thanks guys i just downloaded plague's new workflow and it's gone now, thanks for both of you.

@necroryona

no expert but followed PlagueKind workflow and it fixed my error, if it the correct way? not sure but it did remove the error:

  1. create folder: h3_adaln in "ComfyUI/models" folder
  2. https://huggingface.co/deAPI-ai/minimax-h3-33b-int8/resolve/main/loras/h3_silu_temb_grid.safetensors put this file in h3_adaln folder
  3. image

put this node after your custom lora loader node(if using)

I tried this with experimental lora /minimax_h3_fl2v_lightx2v_turbo_4to8step_v0.1-v1.0_768p_v4_step600_dareties_fro095.safetensors
and the result is terrible. Does it not work?

@Rkss
yes im having trouble with this lora also - if using with custom lora , the elements of the custom lora is deformed ... the light x2v 1.1 fl2va lora keep them correctly, maybe im doing something wrong

@Rkss
yes im having trouble with this lora also - if using with custom lora , the elements of the custom lora is deformed ... the light x2v 1.1 fl2va lora keep them correctly, maybe im doing something wrong

https://github.com/Rkkss/ComfyUI-H3-Turbo-LoRA-Bridge

Well, i vibecoded a loader for this lora . Expect some buggy mess, but it works for now. Or wait, someone to prune it

@Rkss
yes im having trouble with this lora also - if using with custom lora , the elements of the custom lora is deformed ... the light x2v 1.1 fl2va lora keep them correctly, maybe im doing something wrong

https://github.com/Rkkss/ComfyUI-H3-Turbo-LoRA-Bridge

Well, i vibecoded a loader for this lora . Expect some buggy mess, but it works for now. Or wait, someone to prune it

this is very good, thanks! i had to install this node to make it work :) https://github.com/Larryvrh/ComfyUI-MiniMax-H3-Turbo

https://huggingface.co/Rkss/Minimax_h3_fl2v_lightx2v_turbo_4to8step_v0.1-v1.0_768p_v4_step600_dareties_fro095_pruned

Pruned Adaln modules. It works with the native LoRA loader
I tested a bit and couldn't determine if these modules made visible differences
It's not 1:1 with the original LoRA, just so you know.

Sign up or log in to comment