ComfyUI Tutorial : How To Create Consistent Images Using Flux Model

[@DeepImagesAI]

Great video, great workflow and thank you for including the graph.

[@Ozstudiosio]

Hi thanks and really great video, but which part or node exactly I can replace to allow me use my image character to create consistent character sheet too, and also if can give me your contact to speak about some work?

[@p_p]

i think you can add an ultimate sd upscale node to last step and improve the quality 10 times. thanks for sharing

[@cgpixel6745]

yes you can do it i had the same idea but since it takes to much time for my pc to handle it so i did not include

[@DavletshinArtur]

cool workflow, thanks!

[@DesignDesigns]

Thank you so much for the details video....

[@cgpixel6745]

your welcome i am happy that you like it

[@猫见樱]

so cool! very thanks! ♡

[@猫见樱]

I have a question, can face fix & body fix work on 1 person image?

[@cgpixel6745]

Yes it can work

[@猫见樱]

@@cgpixel6745 thank you! I`ll try it!

[@FunniestCatsandPets]

Thanks for the great video and workflow. Can we apply an image prompt instead of a text prompt in order to generate the character sheet of an existing character...I have looked at workflows from other channels as well but could not fınd it. I dont want to go the old way of fooocus because no matter the weights, it still changes the character. I want the consistency that is why I am asking. Thank you for your reply in advance :)

[@cgpixel6745]

yes you can do that by replacing the first part of the workflow with load image node

[@FunniestCatsandPets]

@ thanks a lot 😄

[@devnull_]

Thanks!

[@phenix5609]

i assume with the final rendering image we can use it to create a lora with flux gym, to then being able to generate consistant character with flux in any pose or expression right ? never try creating a lora myself not sure if this is enough for it. ?

[@cgpixel6745]

Yes from what I know it can work but you will need to have at minimum 20 images of your subject

[@RoN43wwq]

great

[@cgpixel6745]

thanks

[@stevenls9781]

is it possible to start with an image that is the poses for full body but also head facing left, right, front and back. Thinking it would be good to train a Lora with body and head like this.

[@cgpixel6745]

yes it is possible to do that by replacing the first part of workflow with load image nodes. it is good way for training lora using this multi pose images

[@stevenls9781]

​@@cgpixel6745solid, thanks for the reply. I am a bit of a hack with all of this but will have a crack.

[@dreamcatcher973]

Hey bro! Thanks for you sharing! I have an issue - save image node inside character sheet is always black (no image generated). Any thoughts?

[@cgpixel6745]

Make sure to use the right vae for the vae decode don't use sdxl vae that comes with sdxl model into flux nodes

[@guijiao]

[@loquacious1956]

I contiually get no vae connected error ... "this controlnet needs a vae" if I connect the VAE from the load checkpoint "juggernautxl" I then get another error "Mat1 and Mat2 shapes cannot be multiplied? Not sure where to go with this now????

[@cgpixel6745]

then just plug the vae from checkpoint loader of the SDXL model it should fix it

[@loquacious1956]

@@cgpixel6745 Every SDXL model I tried from the checkpoint loader gives a Mat1 and Mat2 shapes cannot be multiplied error. I was using the exact model you have in your workflow and it asks for a vae connection if I connect it from the checkpoint loader to the vae on the apply control net I get the same Mat1 and Mat2 shapes cannot be multiplied error

[@yinghaowang1878]

​@@loquacious1956 I encountered a similar issue, and my solution was: in the first work area under the ControlNet model, use the file recommended by the author instead of using Flux's ControlNet.

[@mostafamostafa-fi7kr]

the images have not much quality . can it be fixed ?

[@cgpixel6745]

for SDXL you can use turbo or lightning model as for the final results it should be good since we are using flux model combined with realism lora

[@kevinbatdorf]

This isn’t really a tutorial. You’re just explaining the workflow. Do you have any videos that teach how everything works and fits together?

[@cgpixel6745]

Yeah and when you understand how the workflow work you just have to click queue prompt and it will work automatically so it is a tutorial.if you wanna understand how each nodes works it gonna be very difficult for you.

[@kevinbatdorf]

This is more like a walkthrough of a workflow. No one learns anything here. Anyway, no need to be defense. Take the feedback or ignore it. Up to you.

[@cgpixel6745]

@@kevinbatdorf this video allow you to start from nothing and get an image with multiple angles and consistent results so for me it's enough. Edit: don't bother with the response since I am very convinced with my ideas.

[@kratuskratus1669]

Very good video, however I'm getting this error "ValueError: Fast download using 'hf_transfer' is enabled (HF_HUB_ENABLE_HF_TRANSFER=1) but 'hf_transfer' package is not available in your environment. Try `pip install hf_transfer`." Do you have any tips?

[@cgpixel6745]

from what i know it is related to your internet connexion you can check this link for more info github.com/huggingface/huggingface_hub/issues/1831

[@LUCKYRABITT]

How much ram is needed

[@rivaldofreitasdecarvalho9217]

[Errno 2] No such file or directory: 'D:\\ComfyUI_windows_portable_nvidia\\ComfyUI_windows_portable\\ComfyUI\\custom_nodes\\ComfyUI-tbox\\..\\..\\models\\annotator\\lllyasviel\\Annotators\\.cache\\huggingface\\download\\body_pose_model.pth.25a948c16078b0f08e236bda51a385d855ef4c153598947c28c0d47ed94bb746.incomplete