search

ssube / onnx-web

web UI for GPU-accelerated ONNX pipelines like Stable Diffusion, even on Windows and AMD
https://discord.gg/7CdQmutGuw
MIT License
205 stars 27 forks source link

Documentation: setup-guide.md missing steps/information #451

Open h3ndrik opened 11 months ago

h3ndrik commented 11 months ago

There are two omissions when following the procedure outlined in docs/setup-guide.md

  1. "Test the models": At this point there is no model downloaded and the script fails. It would be nice to add a wget/curl command with the URL to download the specific model that is used by scripts/test-diffusers.py
  2. After "Download the web UI bundle": The instructions abruptly stop at that point. (And are followed by some windows specifics.) Usually guides follow through until the software is running. Perhaps we could add the command that downloads and converts all the models and the command that starts Flask. At least some info on where in the documentation to continue.

And while copying the index.html etc it's not too clear that you need to use the 'api/gui' directory and not the other gui directory.

ssube commented 10 months ago
  1. You are correct, that is a little bit out of order now. You definitely need at least one model to test, and I recently added some pre-converted ones for the Windows bundle (to save time), so I can reuse those and possibly even have the test-diffusers.py script check for the model.
  2. I can add a few more steps, that should not be a problem. At the very least I will go through the launch scripts and then end with a link to the full user guide for adding more models.
  3. The launch scripts will check for the web UI files now, and the powershell launch script will even attempt to download them, but I can make that more clear in the docs as well.
h3ndrik commented 10 months ago

Oh thanks, that makes it straightforward. You even mentioned the api url and everything. Thank you very much. I think a user like me is now set to follow the procedure without looking up additional information. I think we can close this issue.