ComfyUI Tutorial Series: Ep21 - How to Use OmniGen in ComfyUI

[@pixaroma]

UPDATE: the node has been Updated, check the discord server for workflow updates, now it has presets and automatic models and memory management so you let it download the model is best for your pc. Free workflows are available on the Pixaroma Discord server in the pixaroma-workflows channel discord.gg/gggpkVgBf3 You can now support the channel and unlock exclusive perks by becoming a member: pixaroma kzbin.info/door/mMbwA-s3GZDKVzGZ-kPwaQjoin Check my other channels: www.youtube.com/@altflux www.youtube.com/@AI2Play

[@SebAnt]

Thanks - Had been waiting for your version on this !!!

[@pixaroma]

Thank you so much Sebant 🙂

[@59Marcel]

Another fantastic tutorial. I found the work flow disruptions fixes very useful. Thank you.

[@pixaroma]

thank you Marcel 🙂

[@SumoBundle]

Thank you so much for this detailed tutorial on using OmniGen in ComfyUI I really appreciate the effort you put into explaining everything so clearly.

[@ivo_tm]

Fantastic tutorial!

[@pixaroma]

thanks ivo 🙂

[@paul-rk1nz]

Thank you for the tutorials. What model do you recommend for maintaining a consistent character across different situations using ComfyUI? And what do you think about "STOIQO NewReality"?

[@pixaroma]

The best way is training a lora model for a character with different images of that character. I didn't try the models you mentioned

[@paul-rk1nz]

@@pixaroma Thank you

[@baheth3elmy16]

Thank you very much!

[@zorayanuthar9289]

Your channel is fantastic-please keep up the great work! I find your videos incredibly helpful. Could you create content about running ComfyUI in the cloud? Not everyone can invest in an RTX 4090, so exploring cloud-based alternatives would be valuable. Perhaps you could benchmark various cloud providers for ComfyUI performance. I believe many viewers would be interested in your findings.​​ ❤❤❤

[@pixaroma]

Running in cloud cost money and I have an rtx4090 so didn't need a cloud option, so maybe later if my channel grows bigger, right now dont really worth it to spend money just to test

[@newwen2102]

what are the advantages of using omnigen over your previous tutorial which used flux?

[@pixaroma]

I still use flux, this had potential in the future because it doesn't need control net, ip adapters all those things you just prompt for them, just quality is not compared to flux yet

[@hannibal911007]

Still no thumbnails of women with big b**bs on this channel... hum probably because the guy behind it is someone serious and the content is super well done. Thanks man! May I suggest you something in relation with IPAdapter or any equivalent workflow related to this subject, applied to Flux on comfyui.

[@pixaroma]

I have been avoiding any ipadapter, face id, roop etc because of insight face that causes problems, and for some youtubers caused also copyright strike :(

[@larryvw1]

thank you, if I use the cumfyui node do I need to install the omnigen seperatly?

[@pixaroma]

no, it will download anything it need to run on comfyui

[@larryvw1]

Where are the models or safetensoners stored in OmniGen it apears that when using cumfyui it downloads the each time, please advise

[@pixaroma]

it should download only first time, location is ComfyUI_windows_portable\ComfyUI\models\LLM\OmniGen-v1

[@shuntera]

That Omnigen node has changed, has a bunch of other parameters including a couple of different things, like a preset_prompt. Took ages to work myself around it. I can generate the 20 year old, but the image never gets passed to the 2nd Omnigen node which shows up in a red box - looks virtually the same as the first node so not sure why it doesn't work. Maybe you could upload your workflow to Discord after updating your nodes and getting it to work. UPDATE: Found it had to reselect the correct preset prompt from the given list. Not sure how you can just bypass the preset prompt - maybe by choosing None. Might try that later

[@pixaroma]

I just update but it didnt change the node interface for me, it looks the same, hmm

[@Supreeth0112]

I am getting the following error - "RuntimeError: Failed to import transformers.models.phi3.configuration_phi3 because of the following error (look up to see its traceback): No module named 'transformers.models.phi3.configuration_phi3'" Any idea what i need to do?

[@pixaroma]

Did you try to use update dependencies and update comfyui bat from update folder? Try to see if on their page can find a similar issues like i did in the video

[@CsokaErno]

I gave up now. I keep getting error messages. I managed to resolve them somehow, but I can't fix it: "ailab_OmniGen object of type 'NoneType' has no len()"

[@pixaroma]

you can try to use this installer for comfyui and has all the nodes I used ready installed github.com/Tavris1/ComfyUI-Easy-Install or you can check the issues to see if someone has the same error github.com/1038lab/ComfyUI-OmniGen/issues

[@devnull_]

Thanks, really good overview. TBH, personally I don't have time or energy to test yet another model, even if it uses some clever system to understand how to execute different operations based on natural language, instead of the usually require setting up of some extensions or workflows. Image quality is not there yet it seems. Also, not sure if it is appealing to me that my power to tweak settings as a user is taken away as I like to always tweak settings, now these operations resemble more like creating a prompt and hoping the model is understanding what I tried to do, power users don't like it for sure, but maybe this would be more suitable for some Adobe/Bing-like simplified gen AI service.

[@pixaroma]

I am sure the quality will improve in time, if they can combine flux with omingen and sd 3.5 we could get a nice model 😀 but i think the technology is heading that way, where we talk and explain the Ai what we need and deliver, a personal Jarvis

[@devnull_]

@@pixaroma yes, that is why I'm waiting :D - I'm definitely too expecting it to get to point where one can generate images and video, just like they command holodeck in Star Trek, computer will understand the context and all the nuances exactly.

[@GuysLightburnLaserGuides]

I'm starting to think you're AI with all this work you put out bro

[@pixaroma]

😄 thanks

[@bcgd5059]

Can you please do a tutorial on how to install ipadapter?

[@pixaroma]

I am not doing from different reasons, first some of the functions need insightface that cause a lot of trouble on different system when is installed, in the sense that need certain dependencies that might work on my pc and not work on yours, the second reason is that insightface is for personal use only and some KZbinrs got copyright strike from insigntface, but i am sure that will have some alternative in the future without all those problems

[@fabiotgarcia2]

Does it work on Mac?

[@pixaroma]

I don't have a mac to test it sorry, i don't know

[@chesthoIe]

1:40 Mine was "..\..\..\..\Assets\python310\python.exe -m pip install -r requirements.txt" cause I installed Comfy with Stability Matrix. I'm leaving it here in case anyone else is in the same boat. If you get the error: "The system cannot find the path specified." You gotta fix the path to point to the python install your Comfy uses.

[@pixaroma]

thanks, not sure why on stability matrix is different, but some things are different compared with comfyui regular install without stability

[@dioputra2893]

hi can i know python_embeded python version u use please :)

[@pixaroma]

Python version: 3.11.8

[@dioputra2893]

@ thank you

[@Uday_अK]

Alien language 😂

[@pixaroma]

😂

[@elidelia2653]

Omnigen has some cute tricks but IMHO it's overated. Im going to circle back in a couple of months and see if it gets better in performance and quality.

[@pixaroma]

I am sure it will get better, is not much now, just like sd 1.5 wasn't much when was released, but it has grown so much since then :) So we will see