Docker image to run Fooocus on Vast.ai.
What's included in this image?
Below are instructions on how to run this Docker image on Vast.ai
Click on the template link to view it on Vast.ai
Both Fooocus and Filebrowser are protected by credentials by default. The username is admin.
Make sure to change your password in template by editing LOGIN_PASSWORD environment variable passed to Docker on start!
Example LOGIN_PASSWORD=sUp3r$3kr3t!
There is a custom Fooocus preset called next that uses JuggernautXL V9 model. If you want to start Foocus with this preset change FOOOCUS_PRESET environment variable to next.
Example FOOOCUS_PRESET=next
Other valid presets are anime, lcm, lightning, realistic, sai or default
Find an instance close to your location (filter by continent) so you get a low ping.
Filter by GPU. If you just want to play around RTX3090 is enough.
Launch the instance and go to Instances in Vast.ai.
The Docker image is large (7GB) so it will take around 7-8 minutes to start the instance.
When the instance is up (when you see >_CONNECT button) click on the instance IP address on the top and check what ports 3000 and 4000 are mapped to.
Filebrowser listens on port 4000 and Fooocus on port 3000.
Copy the instance IP together with the port and paste it into your browser.
Start with the Filebrowser because Foocus needs longer time to start (it needs to download models on upstart).
Once you log in to Filebrowser you will see a logs directory. In there you will file a foooocus.log file. Open it to see the progress of the model downloads.
When you see in the log file that JuggernautXL (or some other model) is downloaded you can finally login to Fooocus (port: 3000) in your browser.
When you are done with you work you can either destroy or stop the instance (so you don't waste your money).
No. Make sure to download it using Filebrowser (or scp if you are 1337).
It can be one of two. Something went wrong during instance creation or Fooocus is still loading. Check logs/fooocus.log file in the Filebrowser for debugging.
Probably not but I haven't try.
- Build Docker images using GH actions
- Create multiple Fooocus version images and Cuda versions
- Add custom provision script
- Put Fooocus behind a reverse proxy
MIT