Create Consistent Characters in ComfyUI with ControlNet and IPAdapter! 🚀
ฝัง
- เผยแพร่เมื่อ 6 ก.ย. 2024
- Welcome back, everyone! In this video, we're diving deep into the world of character creation with SDXL.
For the installation process and the latest FaceID models, make sure to check out my previous videos and this link: github.com/cub...
Here's the workflow link: comfy.icu/c/h6...
A big shoutout to Matteo aka Latent Vision for his fantastic work on IPAdapter. You can explore more of his insights on his channel: / @latentvision ! 🚀
Thank you! So much potential with this workflow. Have been getting into ComfyUI and your videos are the best.
keep them coming, these are gold!
This is the most logical consistent character workflow Ive seen. Thanks, keep them coming! 🙌🙌🙌
Thank you for this! I look forward to checking out this workflow!
workflow is not available
Hi!
I love this! I saw the original way back when, but this is a nice adaptation to SDXL. I do have a couple questions:
1) Dow well does it function with different openpose models, when seen from the side, standing in weird ways etc.?
2) Can you add features to the face, like sometimes have glasses, sometimes not, but keep those consistent? I.e. If I had a pair of red glasses, could we keep that totally consistent between different characters, and add and remove them at will?
Thanks so much!
1. Ok, the facedetailer might be an issue when facing from the side as it will try to force a frontfacing face. So id recommend to disable it in that case.
2. With some additional steps , i think it can be done. However, the facedetailer doesn’t like sunglasses so you ll have to disable that also.
Insane
Pure gold, keep these please!
Bless, very admirable of you.
that is amazing
Great ! Can you please repost the link plz, the link is dead 😊
This is by far the most comprehensive video on creating a consistent character. Thank you. Liked & Subscribed.
Wow, the workflow is also downloadable. Thanks again!
You re welcome ! 🙏
where did you get ipadapter_clipvision_sd1.5.safetensors from? for the load clipvision node? I can't find it anywhere
One of the nodes doesn't exist any more, and even in the older version (for SD1.5) ComfyUI keeps telling me that I need to install 'ComfyUI IPAdapter Plus', but it's already installed. Both versions have insurmountable obstacles. :(
Hi and good job. This is exactly what I was trying to do but it's still difficult for a beginner like me.
That said, I have a problem, I can't DL your workflow via your link:
No output
No output was found for this run. Please use at least one "SaveImage" node instead of "PreviewImage"
And some nodes in your ComfyUI Graph are red.
Thanks for your sharing! But the workflow is not available.
workflow link is down, 404 error.
Hey, I`m quite new to stable diffusion and image generation in general and learning a lot through your videos. I tried to use your workflow and play with it, but it takes for ever to generate any images with it. The whole process takes around 15-20 min per Image. I already set the batch size to 1. I have a RTX 2070 super (8gbvram) and 16gb ram.
Is there anything qrong with my setup?
Hi great video!.I can't download the workflow,do you have a link that i can download this workflow?
Thanks, but why i have this problem?:
When loading the graph, the following node types were not found:
ControlNetLoader
EmptyLatentImage
SetLatentNoiseMask
LoraLoader
CLIPTextEncode
Nodes that have failed to load will show as red on the graph.
Cause youre missing nodes. Open up the comfyui manager and click on “install missing nodes”
@@tech_bytebrain thanks lot, i change the form that i had installed the ComfyUI, now i have another problem, Where can i find the "IPadapter_clipvision_SD1.5.safetensors" and, in what folder i should put this?
Why does your DW Pose Estimator have 2 more node controls the downloaded version? I'm getting an error at that node (31)
Not sure what you mean . Hit me up on discord discord.gg/MaWuyCDN
Its great until the final render. Mine goes bust using that much controlnet with 6gb vram and the process wont complete.
Perhaps replace the finals upscalers with NN upscaler or do it in a separate step. Hope that makes sense. Edit: rundiffusion is not bad if you want to use comfy online
Hmm, is there a way to download this to run locally? I clicked the link but it shows my the Comfy workflow with no way to download (unless I'm missing something)
hmm the UI has clearly changed. What you can do is click "run workflow" , once it opens you can press "save" in comfyui UI and save the JSON on your PC.
When loading the graph, the following node types were not found:
IPAdapterApply
IPAdapterApplyFaceID
InsightFaceLoader
please
help
IP Adapter has been reworked and you'll have to adapt all workflows to the new nodes 😢
@@tech_bytebrain insightfaceloader does appear for me, but is there any workaround for the IP nodes?
could you use this to make images for a lora? I haven't tried making one yet, are there any good, up to date videos that you know of? thank you!
You can totally use a similar method to create images for a Lora.
What about instantid ? What thw difference
Instantid is more suited for stylising from what ive experienced. It is possible to do realistic portrait but they look a bit burned imo.