"Use all" should include all generation metadata

[Malrama]

Is there an existing issue for this?

• [X] I have searched the existing issues

OS

Windows

GPU

cuda

VRAM

12GB

What version did you experience this issue on?

3.0.1post3 & 3.0.2a

What happened?

I tested 3.0.1post3 and 3.0.2a and just would like to report the following observations regarding metadata and "Use all":

3.0.1post3 and 3.0.2a:

• Both versions do not work properly with "Use all". It seems to only retrieve certain metadata and not everything that would be available to "Use". Missing components are "LORAs and VAEs".
• The following components are not saved in the metadata, thus can't be applied when using "Use all": "Dynamic Prompts and Noise". Since they can change a picture dramatically, it would be nice to get this setting too.

Here is my argument why that would be important: Since it's not hard to have 1000+ pictures in your Galary, it's nearly impossible to remember all the settings used to generated them. I often scroll back to my previous pictures and maybe decide to pick up from there. Being able to trust "Use all" to generate me the same picture (in an optimal case) would be really helpful. It is possible to look into the metadata to get the information about "LORAs and VAEs" but even that step involves some searching and knowledge (reading metadata) that I would love to trust "Use all" for.

3.0.2a

• It currently does not set the model to the used model when "Use all" or "Use model" is selected. It just keeps whatever model is currently selected in the UI.

I hope this Bug report is useful. If there are any further questions or testing needed on my side, I would be glad to assist.

Screenshots

No response

Additional context

No response

Contact Details

No response

[gsreeves]

Metadata isn't written correctly to image in the first place. Perhaps I should open an issue for that. Here are two issues with metadata written to images:

1. Sometimes positive prompt appears in positive style: instead of in prompt:
2. Combinatorial combinations specific to an image are not written, just the whole recipe for all the images, so you can't tell which is which. On the other hand, the data in the database itself looks fine. If program is getting it from db, then carry on.

[dfl]

I also noticed this issue. Also, I don't see any metadata in my downloaded images.

[psychedelicious]

We'd be happy for a contribution for this.

To help you get started, the functions to recall parameters are in invokeai/frontend/web/src/features/parameters/hooks/useRecallParameters.ts.

To recall things like LoRAs, we need to first verify that the models exist. The logic would flow like this:

• Use the RTK Query cache to check if the models are available on the server
• If they are, use one of the existing zod schemas to validate the LoRA or main model or whatever it is you want to recall
• If taht validation passes, dispatch the redux action to update it

Note that the current logic does not validate the model against the query cache! This should be fixed at the same time as recalling LoRAs etc.

You'd need to use the hooks exported from invokeai/frontend/web/src/services/api/endpoints/models.ts to get the available models, and confirm that the model to be recalled is in that list.

See invokeai/frontend/web/src/features/parameters/components/Parameters/MainModel/ParamMainModelSelect.tsx for an example of how to use the hooks. A more refined way to use the hooks is to use selectFromResult, which you can see in action here: invokeai/frontend/web/src/features/ui/components/tabs/ModelManager/subpanels/ModelManagerPanel/ModelList.tsx

If you are familiar with react/frontend dev and would like to take this on, I'm happy to help out. Ping me here or on discord (same username).