Closed fengyuzzz closed 3 months ago
Thanks for your contribution.
Just wondering why the clip text encode nodes don't use the clip from resadapter node? Any particular reason?
I get different result with dreamshaper lcm Normal Render looks better than ResAdapter render but ResAdapter sticks more to the prompt
This is the prompt for the last image :: | bare shoulder,sexy, cleavage,1 girl,portrait, curly hair,simple background, (masterpiece,best quality),lips, spark ,niji style, camisole, Pullover Sweater , realistic,in the villa, steamy, steam , sad,deep red gradient background, |
---|
:: First Image default prompt
Just wondering why the clip text encode nodes don't use the clip from resadapter node? Any particular reason?
If you only use resadapter loader, you need to use clip. But we use resadapter and baseline, so I directly use baseline clip information.
I get different result with dreamshaper lcm Normal Render looks better than ResAdapter render but ResAdapter sticks more to the prompt
I get different result with dreamshaper lcm Normal Render looks better than ResAdapter render but ResAdapter sticks more to the prompt
Can you share your workflow, i try to reproduce it. However, we find that resadapter is not compatible with accelerate lora in resolution extrapolation. Because resadapter is trained in fine-tuning. I am considering to train resadapter with lcm, asd, tcd and so on. Now, i am training new resadapter for broading ratio.
只是想知道为什么剪辑文本编码节点不使用来自resadapter节点的剪辑?有什么特别的原因吗?
如果只使用resadapter loader,则需要使用clip。但是我们使用resadapter和baseline,所以我直接使用基线剪辑信息。
Just wondering why the clip text encode nodes don't use the clip from resadapter node? Any particular reason?
If you only use resadapter loader, you need to use clip. But we use resadapter and baseline, so I directly use baseline clip information.
I don't know what you mean by that. Can you be more specific
The baseline worflow and resadapter workflow share the same text clip embedding. However, there are bugs in this workflow with ip-adapter, controlnet and lcm-lora. I am fixing these bugs.
Thank you for your help
We have fixing model loading bugs, providing more workflows. Detailed information is here.
And there are resadapter_v2 in huggingface.
I provide a demo video, which is helpful for others to use ComfyUI-ResAdapter. @jiaxiangc
https://github.com/jiaxiangc/ComfyUI-ResAdapter/assets/162297627/adc6ec6d-bc43-4ecc-9d3a-2b3c31aeef6c