## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [ ] Optimization
- [X ] Documentation Update
## Have you discussed this change with the InvokeAI team?
- [ X] Yes
- [ ] No, because:
## Description
This PR points mkdocs to the `main` branch again, so that the 3.0.0
documentation appears in gh-pages.
It also makes a minor tweak to the tooltip for model imports, so that
users know that URLs are accepted.
Also rebuilds frontend for use in beta testing.
I've opted to leave out any additional upscaling parameters like scale
and denoising strength, which, from my review of the ESRGAN code, don't
do much:
- scale just resizes the image using CV2 after the AI upscaling, so
that's not particularly useful
- denoising strength is only valid for one class of model, which we are
no longer supporting
If there is demand, we can implement output size/scale UI and handle it
by passing the upscaled image to that a resize/scale node.
I also understand we previously had some functionality to blend the
upscaled image with the original. If that is desired, we would need to
implement that as a node that we can pass the upscaled image to.
Demo:
https://github.com/invoke-ai/InvokeAI/assets/4822129/32eee615-62a1-40ce-a183-87e7d935fbf1
---
[feat(nodes): add RealESRGAN_x2plus.pth, update upscale
nodes](dbc256c5b4)
- add `RealESRGAN_x2plus.pth` model to installer @lstein
- add `RealESRGAN_x2plus.pth` to `realesrgan` node
- rename `RealESRGAN` to `ESRGAN` in nodes
- make `scale_factor` optional in `img_scale` node
[feat(ui): restore ad-hoc
upscaling](b3fd29e5ad)
- remove face restoration entirely
- add dropdown for ESRGAN model select
- add ad-hoc upscaling graph and workflow
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [ ] No, because:
## Description
## Related Tickets & Documents
<!--
For pull requests that relate or close an issue, please include them
below.
For example having the text: "closes #1234" would connect the current
pull
request to issue 1234. And when we merge the pull request, Github will
automatically close the issue.
-->
- Related Issue #
- Closes #
## QA Instructions, Screenshots, Recordings
<!--
Please provide steps on how to test changes, any hardware or
software specifications as well as any other pertinent information.
-->
## Added/updated tests?
- [ ] Yes
- [ ] No : _please replace this line with details on why tests
have not been included_
## [optional] Are there any post deployment tasks we need to perform?
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [ ] Bug Fix
- [x] Optimization
- [ ] Documentation Update
## Have you discussed this change with the InvokeAI team?
- [x] Yes
- [ ] No, because:
## Description
There no vram cleanup on models offload which leads to filling vram and
slow generation speed.
## What type of PR is this? (check all applicable)
- [x] Feature
- [x] Optimization
## Have you discussed this change with the InvokeAI team?
- [ ] Yes
- [ ] No, because:
## Description
## Related Tickets & Documents
<!--
For pull requests that relate or close an issue, please include them
below.
For example having the text: "closes #1234" would connect the current
pull
request to issue 1234. And when we merge the pull request, Github will
automatically close the issue.
-->
- Related Issue #
- Closes #
## QA Instructions, Screenshots, Recordings
<!--
Please provide steps on how to test changes, any hardware or
software specifications as well as any other pertinent information.
-->
## Added/updated tests?
- [ ] Yes
- [ ] No : _please replace this line with details on why tests
have not been included_
## [optional] Are there any post deployment tasks we need to perform?
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [ ] Feature
- [x] Bug Fix
- [x] Optimization
- [ ] Documentation Update
## Description
Various fixes to consume less memory and make run sdxl on 8gb vram.
Most changes due to moving all output tensors to cpu, so that cached
tensors not consume vram.
