## What type of PR is this? (check all applicable)
- [ ] Refactor
- [x] Feature
- [ ] Bug Fix
- [x] Optimization
- [x] Documentation Update
- [ ] Community Node Submission
## Have you discussed this change with the InvokeAI team?
- [x] Yes
- [ ] No, because:
## Have you updated all relevant documentation?
- [x] Yes
- [ ] No
## Description
- updates the Docker image with ubuntu23.04 base, python3.11
- use the newer pytorch wheel with cuda12.1 support
- corrects `docker compose` CLI in shell script wrappers and docs
- update / overhaul Docker docs
- clean up obsolete lines in `.gitignore`
## QA Instructions, Screenshots, Recordings
Follow the documentation changes, or simply:
```bash
cd docker
cp .env.sample .env
# Set your INVOKEAI_ROOT in .env
docker compose up
```
## Added/updated tests?
- [ ] Yes
- [x] No : N/A
Custom nodes may be places in `$INVOKEAI_ROOT/nodes/` (configurable with `custom_nodes_dir` option).
On app startup, an `__init__.py` is copied into the custom nodes dir, which recursively loads all python files in the directory as modules (files starting with `_` are ignored). The custom nodes dir is now a python module itself.
When we `from invocations import *` to load init all invocations, we load the custom nodes dir, registering all custom nodes.
Also added config options for metadata and workflow debounce times (`metadataFetchDebounce` & `workflowFetchDebounce`).
Falls back to 0 if not provided.
In OSS, because we have no major latency concerns, the debounce is 0. But in other environments, it may be desirable to set this to something like 300ms.
- Refactor how metadata is handled to support a user-defined metadata in graphs
- Update workflow embed handling
- Update UI to work with these changes
- Update tests to support metadata/workflow changes
This fixes a weird issue where the list images method needed to handle `None` for its `limit` and `offset` arguments, in order to get a count of all intermediates.
On our local installs this will be a very minor change. For those running on remote servers, load times should be slightly improved.
It's a small change but I think correct.
This should prevent `index.html` from *ever* being cached, so UIs will never be out of date.
Minor organisation to accomodate this.
Deleting old unused files from the early days
## What type of PR is this? (check all applicable)
- [ ] Refactor
- [x] Feature
- [ ] Bug Fix
- [ ] Optimization
- [ ] Documentation Update
- [ ] Community Node Submission
## Description
This PR adds the ability to pass multiple images to a single IP-Adapter
(note the difference from using _multiple IP-Adapters at once_, which is
already supported.). The image embeddings are combined in the IP-Adapter
attention layers. This is the same strategy for combining multiple
images as used in Insta-LoRA workflows
(https://civitai.com/articles/2345).
This PR only adds multi-image support in the backend and the node
editor. The Linear UI still needs to be updated.
## QA Instructions, Screenshots, Recordings
I have manually tested the following via the workflow editor:
- Multiple images with a single IP-Adapter
- Multiple images per IP-Adapter, and multiple IP-Adapters
- Both standard and sequential conditioning
- IP-Adapters still work in the Linear UI.
Please hammer at this feature some more with manual testing.
## Added/updated tests?
- [x] Yes
- [ ] No
I updated the existing IP-Adapter smoke test, but it provides pretty
limited coverage of this feature. This feature would probably be best
tested by an end-to-end workflow test, which is not currently supported.
(I'm hoping to put some effort into workflow-level testing soon.)
