1. add a /sync route for synchronizing the in-memory model lists to
models.yaml, the models directory, and the autoimport directories.
2. add optional destination_directories to convert_model and merge_model
operations.
3. add /ckpt_confs route for retrieving known legacy checkpoint configuration
files.
4. add /search route for finding all models in a directory located in the server
filesystem
DONE:
- Restore Update Model functionality
- Restore Delete Model functionality
- Restore Model Convert functionality
- Restore Model Merge functionality
- Refine UX (fine tweaks when everything is done - TODO)
TODO
- Add Model (will be finished in a future PR once the backend work is
done)
IAIMantineSelect and IAIMantineMultiSelect have a bit of extra logic that prevents simple select functionality from working as expected.
- extract the styles into hooks
- rename those two components to IAIMantineSearchableSelect and IAIMantineSearchableMultiSelect
- Create IAIMantineSelect (which is just a dropdown) and use it in model manager and a few other places
When we only have a few options to present and searching is not efficient, we should use this instead.
Image files are immutable and we expect deletion to result in no further
requests for a given image, so we can set the max-age to something
thicc.
Resolves#3426
@ebr @brandonrising @maryhipp
- simplify UI logic in `ModelManagerPanel` components
- fix up the types a bit to make it easier to select models
- remove `openModel` state, just make it a useState since it is very local to model manager
similar to the previous commit, update the node editor to not just store models as strings - instead, store the model object.
the model select components in nodes are now just kinda copy-pastes over the linear UI versions of the same components, but they were different enough that we can't just share them.
i explored adding some props to override the linear ui components' logic, but it was too brittle. so just copy/paste.
We were storing all types of models by their model ID, which is a format like `sd-1/main/deliberate`.
This meant we had to do a lot of extra parsing, because nodes actually wants something like `{base_model: 'sd-1', model_name: 'deliberate'}`.
Some of this parsing was done with zod's error-throwing `parse()` method, and in other places it was done with brittle string parsing.
This commit refactors the state to use the object form of models.
There is still a bit of string parsing done in the to construct the ID from the object form, but it's far less complicated.
Also, the zod parsing is now done using `safeParse()`, which does not throw. This requires a few more conditional checks, but should prevent further crashes.
* feat(ui): salvaged gallery UI enhancements
* restore boardimage functionality, load boardimages and remove some cachine optimizations in the name of data integrity
* fix assets, fix load more params
* jk NOW fix assets, fix load more params
---------
Co-authored-by: Mary Hipp <maryhipp@Marys-MacBook-Air.local>
Co-authored-by: Mary Hipp Rogers <maryhipp@gmail.com>
- available infill methods is server state - remove it from client state, use the query to populate the dropdown
- add listener to ensure the selected infill method is an available one
As it said in comment to this branch we want to use conditioning run:
```python
if cfg_injection: # only applying ControlNet to conditional instead of in unconditioned
```
But in code used unconditioning
embeddings(`conditioning_data.unconditioned_embeddings`).
Later in code confirms that we want to run conditioning generation by
comment and tensor concatenation order(as all code expect to get [uc, c]
tensor):
```python
if cfg_injection:
# Inferred ControlNet only for the conditional batch.
# To apply the output of ControlNet to both the unconditional and conditional batches,
# add 0 to the unconditional batch to keep it unchanged.
down_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_samples]
mid_sample = torch.cat([torch.zeros_like(mid_sample), mid_sample])
```
