8-step lora?
I see that you have several 4-step lightning options. Will you be adding 8-step options? The 8-step version always gave me better results, so I'm hoping you'll make one for this new Qwen Edit—unless you think the 4-step is already great.
Steps 4 and 8 are both applicable
@97Buckeye
Hi,
We will release a 8-step version, before that, you may use 8-step inference for this 4-step model.
@97Buckeye
Hi,
We will release a 8-step version, before that, you may use 8-step inference for this 4-step model.
Hi, is it possible to fix the fp8 model ? Not working on comfyUI
qwen_image_edit_2511_fp8_e4m3fn_scaled_lightning_comfyui.safetensors
+1 , images gets red hue added for some reason with 4 steps , could be just 2511 tho
@siraxe Maybe you are using workflow from 2509. For 2511 you also need to add "FluxKontextMultiReferenceLatentMethod" node to positive and negative conditionings
@97Buckeye
Hi,
We will release a 8-step version, before that, you may use 8-step inference for this 4-step model
I cant find any info regarding this but gpt suggested that if the distill lora been trained for specific steps adding extra steps or lowering the lora strength might be an issue, but i see in community some users lower the lora strength and add few extra steps, can you please confirm on both cases. if its normal to play with strength and add extra steps?
thanks for the 8 steps.. 4 steps is not usable at all its like Q2 gguf 😁🤣