.jpeg)
hello again everyone ,
this is my third article about generating thing on without civit ai , i suggest reading this first This Article About Generating anything since this is a continuation (up to the point where we copied the colab page to our google drive should be enough)
in article above i mentioned how to generate on google colab with comfyui that save your images , checkpoint/loras on google drive wich had a 15gb limit. here we focus on how to download large models (like flux and models larger than +15gb)
im updating this article in future when i got the time to test and find ways for heavier models if even possible, this is justs barebone version for now just to get the basic things going and showing the method it uses, save it so when update come out you dont miss it.
SAVE YOUR IMAGES TO YOUR DEVICE ! remember this line (it get relevent)
also theres this article about adding loras to your workflow check it out
things in this article :
how to setup comfyui on COLAB (not google drive)
downloading and sorting large models
using large models in ComfyUI
im assuming you already have the colab workspace copied already, i explain it here
i have to reference it alot in this article , also since this is on colab session if your colab session ends the models you downloaded image you made will be deleted (read pros/cons part for explainning)
Introduction :
so in last article we use a google drive to save our checkpoints wich was useful for 1 or 2 checkpoints and maybe 3 4 loras , but what if we want to download larger models ? or more loras?
for that we going to use the resources that google colab alocated for us , its around 70 - 75 GB
Step 1. Install ComfyUI on Colab :
first turn off use google drve checkmark so it dont request access to google drive , this causes comfyui to be installed on colab session only.
After you run the code you should have comfyui inside your colab folder
Step 2. Downloading Checkpoints :
just like how we download checkpoints into google drive we do the same here just go to the checkpoint page and copy the download link
then paste it in the code block that download your models , when you start the code its should downloading your model into the location you write in the second part.
first part of code is where you paste your link between this "" and second part is where you save it in your comfy ui folder
i downloaded Flux Dev - Pruned Model Fp32
i highly suggenst saving your checkpoints/loras anything you download in correct folder.
now we are saving flux into our checkpoint folder here
-O "./models/checkpoints/Flux.safetensors"
if you downloaded a lora accidentally , you can just drag and drop it into the loras folder
or before downloading save it to the right folder like this
-O "./models/Loras/(your lora name).safetensors"Step 3. Generating and Pros/Cons :
after your comfyui url comes just open it and start generating , if you dont know how to get the url just run the last code block where it wrote "Run comfyui" and open the url it give you after few second/minute (if this is a bit unclear check main artice Step 8. Run ComfyUi part)
there is no workflow? check this article
im not a big flux user so i dont know the best settings so do your thing, now when you generate your images it will go to your output folder
!! REMEMBER TO SAVE YOUR IMAGES TO YOUR DEVICE !!
since were using colab session instead of google drive , if your session ends all the images / downloaded model wont stay
Pros & Cons :
here i list a pros and cons of using colab session (Read "Other Things" section)
Pros :
enables using large models like flux
quicker than using google drive method
no limit on prompts you enter
no limit on how many image you generating at same time
large amount of space for different loras/checkpoints
Cons :
since its a colab session , if your session end everything you downloaded/made will be gone
some process heavy models may use too much vram wich stops the "run comfyui" code
setup takes longer than google drive method
have to download checkpoints/loras for each usage
if its a really process heavy model it might take more time to generate things obviously
sometimes it might disconnect if left idle
Other Things :
here i write stuff i couldnt fit in article
1- so i tried using flux full model , it use too much vram wich causes the code to stop , so yes you can download large size model but if the process is too heavy it might stop your generation
2 - ill update this article if i find a way to reduce the vram usage seen some code that limit the vram , save this article so you wont miss out on it
3 - SAVE YOUR IMAGES TO YOUR DEVICE ! as i said its on colab session so save anything you make by either right click and save image or go to comfyui output folder and download images
4 - yes its a bit annoying to have to download the checkpoint each time you want to make something but i dont see any other way without spending money to either upgrading your gpu or buying more storage in google drive
5 -since its 5 hour gpu usage for colab i recommend a generating your images in one sitting so you dont have to keep redownloading a checkpoint again and again
6 - if you got disconnected from the session dont panic first try to reconnect to your colab page if its still there youre all good
if you like the article you can support me by checking my checkpoint here
make some cool stuff :D
my next article probably be about upscaling and highres fix workflows in comfyui Wich make HUGEE difference in your images ,have to see how much work it need ;)
hope this article helped you beautiful creators to generate things with larger models that you like ,
i wish you an amazing week and a good luck !