Metadata-Linked (now part or core as of invoke 5.7) | Invoke | Page 1

willow tendon Dec 21, 2023, 6:05 PM

#

** These nodes have been adopted into core as of invoke 5.7.
I suggest just deleting the metadata-linked-nodes folder and use core insted.
The repo will remain as a source of examples and documentation.**

A set of nodes for Metadata. Collect Metadata from within an iterate node & extract metadata from an image.

Metadata Item Linked - Allows collecting of metadata while within an iterate node with no need for a collect node or conversion to metadata node.
Metadata From Image - Provides Metadata from an image.
Metadata To String - Extracts a String value of a label from metadata.
Metadata To Integer - Extracts an Integer value of a label from metadata.
Metadata To Float - Extracts a Float value of a label from metadata.
Metadata To Scheduler - Extracts a Scheduler value of a label from metadata.
Metadata To Bool - Extracts Bool types from metadata
Metadata To Model - Extracts model types from metadata
Metadata To SDXL Model - Extracts SDXL model types from metadata
Metadata To LoRAs - Extracts Loras from metadata.
Metadata To SDXL LoRAs - Extracts SDXL Loras from metadata
Metadata To ControlNets - Extracts ControNets from metadata
Metadata To IP-Adapters - Extracts IP-Adapters from metadata
Metadata To T2I-Adapters - Extracts T2I-Adapters from metadata
Denoise Latents + Metadata - This is an inherited version of the existing Denoise Latents node but with a metadata input and output.
Flux Denoise + Metadata - Same as above

See repo for instruction on how to install
Repo: metadata-linked-nodes

https://github.com/skunkworxdark/metadata-linked-nodes/blob/main/images/workflow-metadata_linked.png
https://github.com/skunkworxdark/metadata-linked-nodes/blob/main/images/workflow-image_to_metadata.png

GitHub

GitHub - skunkworxdark/metadata-linked-nodes: Metadata nodes for In...

Metadata nodes for InvokeAI. Contribute to skunkworxdark/metadata-linked-nodes development by creating an account on GitHub.

GitHub

metadata-linked-nodes/images/workflow-metadata_linked.png at main ·...

Metadata nodes for InvokeAI. Contribute to skunkworxdark/metadata-linked-nodes development by creating an account on GitHub.

GitHub

metadata-linked-nodes/images/workflow-image_to_metadata.png at main...

Metadata nodes for InvokeAI. Contribute to skunkworxdark/metadata-linked-nodes development by creating an account on GitHub.

willow tendon Dec 21, 2023, 6:56 PM

#

Release History:
27 Feb 25 - All nodes adopted into core invoke as of V5.7

29 Jan 25

added a custom label to the Metadata to LoRA Collection node. Now you can pull and store lora data under any metadata label name.

23 Jan 25

Added flux specific built in metadata names guideance, cfg_scale_start_step & cfg_scale_end_step
Added Automatic recording of flux specific metadata to FluxDenoise node - cfg_scal, cfg_scale_start_step & cfg_scale_end_step
new node Metadata To LoRA Collection - Creates a Lora Collection from metadata that can be fed into the Lora Collection Loader nodes.

28 Sep 24 - invoke 5.0 Flux update. Added Flux Denoise + metadata node. (auto collects model, seed, width, height, steps, guidance, denoise_start, denoise_end and lora metadata.

The previous release is available under the repo tag of invoke-4.2.9

02 Jul 24 - fix to the model identification from metadata. New versions of metadata_to_loras and metadata_to_sdlx_loras

28 Jun 24

Minor fix for DenoiseLatentsInvocation import path introduced with Invoke 4.2.5
New tag to reference Invoke 4.2.4 compatible version

9 May 24 - Updated to use invocation_api - no functional changes

2 Apr 24 - 4.0 Update
The main branch has been updated to be InvokeAi 4.0 compatible and the 4.0-test branch removed. Run the update.bat/update.sh in the node folder to get the latest version.

If you require the 3.7 or older compatible version then there are tags on the repo at the appropriate points that can be used. See the readme in the repo for further details.
26 Feb 24 - Bug fix to Metadata To Model & Metadata To SDXL Model + Revised workflow examples

see repo for full history

21 Dec '23 - First Release

willow tendon Jan 14, 2024, 5:01 PM

#

TODO:

additional flux metadata when added to invoke, VAE, T5 ~Encoder, Clip, Controlnets, ip-adapters etc...

willow tendon Feb 10, 2024, 7:02 PM

#

reserved

willow tendon Feb 15, 2024, 11:06 PM

#

The Release is now available from the Repo

naive dust Feb 16, 2024, 8:53 PM

#

I appear to having issues with Metadata to Model. After genning with Denoise+Meta trying to pull the model data back out I get an error

  File "C:\StableDiffusion\invokeai3\.venv\lib\site-packages\invokeai\app\services\invocation_processor\invocation_processor_default.py", line 134, in __process
    outputs = invocation.invoke_internal(
  File "C:\StableDiffusion\invokeai3\.venv\lib\site-packages\invokeai\app\invocations\baseinvocation.py", line 669, in invoke_internal
    output = self.invoke(context)
  File "C:\StableDiffusion\invokeai3\nodes\metadata-linked-nodes\metadata_linked.py", line 383, in invoke
    model = MainModelField.model_validate_json(
  File "C:\StableDiffusion\invokeai3\.venv\lib\site-packages\pydantic\main.py", line 538, in model_validate_json
    return cls.__pydantic_validator__.validate_json(json_data, strict=strict, context=context)
pydantic_core._pydantic_core.ValidationError: 1 validation error for MainModelField
  JSON input should be string, bytes or bytearray [type=json_type, input_value={'model_name': 'doomerBoo...in', 'submodel': 'unet'}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.6/v/json_type

[2024-02-16 20:43:30,862]::[InvokeAI]::ERROR --> Error while invoking:
1 validation error for MainModelField
  JSON input should be string, bytes or bytearray [type=json_type, input_value={'model_name': 'doomerBoo...in', 'submodel': 'unet'}, input_type=dict]
    For further information visit https://errors.pydantic.dev/2.6/v/json_type```

#

There's also a way to put a separate VAE in to the metadata but no way to pull one out.

willow tendon Feb 16, 2024, 9:01 PM

#

naive dust I appear to having issues with Metadata to Model. After genning with Denoise+Met...

is this a specific model that you're having a problem with or all of them?

#

could you share an image of how you have setup your nodes?

naive dust Feb 16, 2024, 9:04 PM

#

willow tendon is this a specific model that you're having a problem with or all of them?

Appears to be any model. Doesn't matter if I'm pulling the metadata from a loaded image or sticking a Metadata to Model node immediately after a Denoise+Meta node. Does the same with images from t2i too.

willow tendon Feb 16, 2024, 9:05 PM

#

ok I have reproduced. I will take a look at this

willow tendon Feb 16, 2024, 9:30 PM

#

naive dust Appears to be any model. Doesn't matter if I'm pulling the metadata from a loade...

Hopefully that should be fixed now if you pull the repo again

naive dust Feb 16, 2024, 10:40 PM

#

willow tendon Hopefully that should be fixed now if you pull the repo again

That's it. Works great, thanks.

#

Would love to see a Metadata to VAE node.

willow tendon Feb 16, 2024, 10:41 PM

#

np to be fair that bit was obviously untested 🙂

naive dust Feb 16, 2024, 10:41 PM

#

Well it's tested now.

#

Real handy for a simple upscale workflow. Just loras missing, really

willow tendon Feb 16, 2024, 10:42 PM

#

naive dust Would love to see a Metadata to VAE node.

I don't think that should be too hard to do. I will have a look at it in the next day or so. when I get a little free time

willow tendon Feb 17, 2024, 2:13 PM

#

naive dust Would love to see a Metadata to VAE node.

This has now been added

willow tendon Feb 17, 2024, 2:15 PM

#

naive dust Real handy for a simple upscale workflow. Just loras missing, really

Loras are captured but not sure how I could pull them back into a workflow as you could have multiple loras. Maybe it can be done I will have a think about it for a bit

naive dust Feb 17, 2024, 5:00 PM

#

willow tendon Loras are captured but not sure how I could pull them back into a workflow as yo...

Same issue with ControlNets and IP-Adapters I suppose.

willow tendon Feb 17, 2024, 5:02 PM

#

naive dust Same issue with ControlNets and IP-Adapters I suppose.

yes mostly. I think it could be done but would be much easier to break if the image that was used had gone for some reason.

naive dust Feb 17, 2024, 5:03 PM

#

willow tendon yes mostly. I think it could be done but would be much easier to break if the ...

Right but isn't that true of anything in Invoke? A workflow with an image set in it will lose the image if it's deleted, right?

willow tendon Feb 17, 2024, 5:05 PM

#

naive dust Right but isn't that true of anything in Invoke? A workflow with an image set in...

however for controlnet/adapters they tend to use intermediates more because of preprocessors and they can be cleared

#

controlnet/adapters probably could be done a bit like my linked adapters nodes.

naive dust Feb 17, 2024, 5:07 PM

#

Ah, I see.

willow tendon Feb 17, 2024, 5:10 PM

#

However, even with the cleared intermediates issues it is still probably worth having a go at doing them. So on my todo list is: Metadata To Loras, Metadata To Controlnets, Metadata To IP-Adapters, Metadata To T2I-Adapter 😄

naive dust Feb 17, 2024, 5:15 PM

#

They'd be wonderful, though I think I'd only really get use out of the LoRAs. If I was doing CNs or IPAs that'd be a special case which wouldn't be too much of a pain to handle manually.

willow tendon Feb 17, 2024, 5:16 PM

#

I was also thinking that I should have a version of the Metadata to Model that outputs the Unet, Clip and VAE like the existing Main Model nodes does then you don't need the overloaded one from the XYGrid nodes

naive dust Feb 17, 2024, 5:17 PM

#

Is there any use for the full Model output type besides plugging it in to the Core Metadata node?

willow tendon Feb 17, 2024, 5:18 PM

#

yes if you want to feed into an XYGrid

naive dust Feb 17, 2024, 5:18 PM

#

I haven't tried it.

willow tendon Feb 17, 2024, 5:20 PM

#

XYGrids are great for comparing different settings. However, model switching can make it slow especially if they are SDXL models

naive dust Feb 17, 2024, 5:22 PM

#

I just slap some text on images with the composition nodes and throw them in PureRef, I find it easier than panning around one large image.

willow tendon Feb 17, 2024, 5:25 PM

#

not seen pureref

naive dust Feb 17, 2024, 5:26 PM

#

Just an minimalist image viewing board I use for 3D modelling reference but I find it handy for use with AI art stuff too.

naive dust Feb 18, 2024, 10:27 PM

#

Got a chance to play some more and found a little issue in the Metadata to VAE node.
Firstly it will fail with a type error if the VAE metadata is "Default", meaning the node's Default Value dropdown never gets used. This will happen if the image is genned in the t2i tab (while it's set to Default VAE) or with Denoise+meta and no separate VAE is specified using a Metadata Item.
In my use-case, I just want it to use the VAE that's specified in the metadata so perhaps the node should take a "fallback" VAE as an input, instead of the Default Value dropdown. The user can then plug in the main model's baked VAE. If a separate one is specified it will output that, if not it just outputs the fallback.

willow tendon Feb 19, 2024, 11:44 AM

#

naive dust Got a chance to play some more and found a little issue in the Metadata to VAE n...

Will take a look at that

willow tendon Feb 19, 2024, 7:19 PM

#

New Release:

Added Metadata to Loras, Metadata To SDXL LoRAs.
Added unet, clip and vae outputs to Metadata To Model & Metadata To SDXL Model
Added default vae as an input to Metadata to VAE

willow tendon Feb 19, 2024, 7:19 PM

#

naive dust Got a chance to play some more and found a little issue in the Metadata to VAE n...

Done and a few other things. let me know what you think.

#

Still on the todo list Metadata To Controlnets, Metadata To IP-Adapters, Metadata To T2I-Adapter left to

naive dust Feb 19, 2024, 9:20 PM

#

willow tendon Done and a few other things. let me know what you think.

All seems to work fantastically but I did notice that it's odd is that the Denoise+Meta adds a VAE field when that's not part of it's inputs (and it's always Default no matter what you're actually using) and it doesn't add Clip Skip when it seems like, if LoRAs are in it's metadata output, then Clip Skip should be too if it's possible.
Otherwise fantastic work. Huge QoL improvement.

willow tendon Feb 19, 2024, 9:23 PM

#

the vae is pulled from the internal variables within denoise. so if one isn't provided then it comes from the model.

#

I guess I could check if the model name is the same as the main model and omit it but I don't know if the adds anything useful

willow tendon Feb 19, 2024, 9:26 PM

#

naive dust All seems to work fantastically but I did notice that it's odd is that the Denoi...

will take a look at Clipskip. It's not something that I've ever used.

naive dust Feb 19, 2024, 9:34 PM

#

Mostly just the anime models using it, I think but you need it set right to get the expected output of models and LoRAs trained with it.

willow tendon Feb 19, 2024, 10:00 PM

#

I guess at the moment the clip skip could be captured manually. and re-applied via the clip skip node. It looks like the front end only supports clip skip on the 1.5 models

naive dust Feb 19, 2024, 10:06 PM

#

willow tendon I guess at the moment the clip skip could be captured manually. and re-applied ...

Indeed, it can. It just feels like, if the Model and LoRAs get put in to the metadata than the clip skip should too as they're directly related but I'm more than happy as is.

willow tendon Feb 19, 2024, 10:08 PM

#

From what I can see it's not obvious where I could get the clip skip from within the denoise node as it's not specifically passed to it as far as I can see and this happens mostly in the compel node. Also, in nodes you have lots of flexibility of when you would apply clip skip before or after Loras and which clip model etc. from what I can see it's not really a global thing like it is shown in the frontend ui

naive dust Feb 19, 2024, 10:10 PM

#

I don't believe it going before or after LoRAs matters, it still seems to apply correctly but that's purely anecdotal. It's hardly a big issue, though. Thanks for looking in to it.

willow tendon Feb 19, 2024, 10:11 PM

#

np. I might look into it a little more after I have looked at the controlnets and adapters

naive dust Feb 19, 2024, 10:11 PM

#

Oh yea. I'm sure they will be much more desired.

willow tendon Feb 19, 2024, 10:12 PM

#

I feel like the metadata nodes will be in a fairly good place after I have added them

naive dust Feb 19, 2024, 10:14 PM

#

Feels like they're in a great place right now but with CNs and Adapters it'll be the whole package.

willow tendon Feb 19, 2024, 10:20 PM

#

naive dust All seems to work fantastically but I did notice that it's odd is that the Denoi...

just realised what you said here about the vae. it's not the denoise that adds a vae as default its the front end that shows that if one isn't found in the metadata. you have to add vae as a manual metadata field if you want to record it specifically. a bit like this.

#

I was thinking of adding a lattents to image + metadata to capture the last couple of fields automatically but I don't think it is worth the effort.

naive dust Feb 19, 2024, 10:24 PM

#

Ah that makes some sense, I suppose.

#

No, I wouldn't say so. I just used a couple of Metadata Linked nodes at the end to add the few things that Denoise+Meta doesn't grab like ClipSkip. I've got a bit of a node soup going in some places but I'm not sure if that's an issue you should address here or if it's more of a lack of tools in the workflow editor to mitigate that such as groups or reroutes.

willow tendon Feb 19, 2024, 10:28 PM

#

naive dust No, I wouldn't say so. I just used a couple of Metadata Linked nodes at the end ...

If you have some good workflows based on these nodes it would be good to share them in https://discord.com/channels/1020123559063990373/1130291608097661000

#

mine are a spaghetti mess also 🙂

#

naive dust Feb 19, 2024, 10:28 PM

#

I was thinking about it for some more beginner t2i/upscale stuff but I'll have to pull out the other community nodes first.

willow tendon Feb 19, 2024, 10:29 PM

#

you should have seen some of the XYGrid Tiled upscaling workflows 🙂

#

right need to go spend some time with the family. 😄

naive dust Feb 19, 2024, 10:31 PM

#

Great work! ❤️

willow tendon Feb 22, 2024, 4:54 PM

#

New Release: 22nd Feb 2024 - Add New nodes

Metadata To ControlNets - Extracts ControNets from metadata
Metadata To IP-Adapters - Extracts IP-Adapters from metadata
Metadata To T2I-Adapters - Extracts T2I-Adapters from metadata

These nodes have been tested but should be considered as needing more testing for the moment.

Note: Generations will fail if the images that the ControlNet, IP-Adapter or T2I-Apater uses don't exist. As some of the inputs to these can be intermediates then as a user you might not even know that they have been removed.

tired vigil Feb 25, 2024, 7:44 PM

#

Hi, just installed your @willow tendon work and loaded the json "Metadata from Image". All works, can load the image, but it seems to demand a SD 1.5/2 model to be used (XL models are not recognized. I'm using Invoke 3.7). Am I doing something wrong or have I misread the XL readout and it is not supported?

willow tendon Feb 25, 2024, 8:05 PM

#

@tired vigil When you say load the JSON. Do you mean loaded one of the workflows? If so they are a little old now and are only there as a guide of how the nodes could be used. If you want to use that first workflow with sdxl then you would need to change to main model loader and the prompts to sdxl versions. Creating new workflow examples is sontjing that is on my to-do list.

tired vigil Feb 25, 2024, 9:05 PM

#

willow tendon <@224961171420217344> When you say load the JSON. Do you mean loaded one of th...

I see. Yea, I meant I've imported the first of your workflows from the github, which in turn did not seem to support SDXL in its current form. Could have been just me misreading or Invoke 3.7 not supporting it anymore. Thanks for clarifying.

willow tendon Feb 25, 2024, 11:06 PM

#

tired vigil I see. Yea, I meant I've imported the first of your workflows from the github, w...

It should work on invoke 3.7 as that's what I am using however I've not tried reimporting the workflow. I will try and get round to remaking the workflows and do a sd1.5 and sdxl versions however they really are not that different.

tired vigil Feb 26, 2024, 11:08 AM

#

The reason I installed it to begin with was the misconception of Invoke AI being unable to load the meta data of saved images (at least there generated). Turns out that a few examples just had none, cause of how I saved them.
Long story short, I wonder what I'd use it for, now that this has been resolved.

willow tendon Feb 26, 2024, 2:17 PM

#

tired vigil The reason I installed it to begin with was the misconception of Invoke AI being...

well if your still interested I am now working on creating some new workflows. I am also fixing a bug in the metadata to model nodes. this should be done in a few hours

willow tendon Feb 26, 2024, 6:04 PM

#

New Release - 26 Feb 2024

Bug fix to Metadata To Model & Metadata To SDXL Model for when no metadata is found.
Metadata To VAE - label default_vae changed to default_value to match all other nodes
Revised workflow examples and readme

tired vigil Feb 27, 2024, 10:20 AM

#

willow tendon well if your still interested I am now working on creating some new workflows. I...

Can't really make much use of it, since the basic import of meta data was all that I was looking for. I'm just not sure how a node-based meta data viewer would improve my workflow over having it all and deleting/adding things myself.
Just a preference thing, I suppose. But thanks anyways.

balmy orbit Feb 27, 2024, 2:57 PM

#

is there a way to save the positive and negative prompts as well? (models, seed, steps etc. work just fine from "denoise with metadata").

willow tendon Feb 27, 2024, 6:27 PM

#

balmy orbit is there a way to save the positive and negative prompts as well? (models, seed,...

you need to capture the positive and negative with the linked metadata node. a bit like this. Take a look at the workflow examples in the repo if you need an example you can load up.

willow tendon Feb 27, 2024, 6:31 PM

#

tired vigil Can't really make much use of it, since the basic import of meta data was all th...

I am not sure I understand what you mean by "node-based metadata viewer". The nodes allow you to extract metadata from images with metadata and use that to generate new images and store metadata in newly generated images. What is it you are trying to do?

tired vigil Feb 27, 2024, 6:39 PM

#

willow tendon I am not sure I understand what you mean by "node-based metadata viewer". The n...

Yes, but all it does is giving a node to upload an image and splitting the meta data output in their respective categories/nodes, right?
Uploading an image made (and correctly) saved with Invoke does the same, as in putting all the info into their textfields/adjusting values etc.

At least to me, latter seems preferable. Using the node approach would only be better when analyzing non-Invoke (native) gens, or am I misunderstanding how it works?

willow tendon Feb 27, 2024, 6:59 PM

#

yes, that is true if you are only working with one image at a time. but if you then use it with some other kind of node, like images from board. then you can loop over all the images in a board and modify the metadata and regenerate all the image. maybe with a different model or sampler or prompt etc.

balmy orbit Feb 27, 2024, 8:28 PM

#

willow tendon you need to capture the positive and negative with the linked metadata node. a ...

i actually tried this as well based on the examples provided on github. will test it tomorrow again and let you know how it went afterward.

tired vigil Feb 27, 2024, 8:34 PM

#

willow tendon yes, that is true if you are only working with one image at a time. but if you ...

Thanks for the explanation. It definitely seems useful for power-users, but not for me currently.

balmy orbit Feb 27, 2024, 10:37 PM

#

balmy orbit i actually tried this as well based on the examples provided on github. will tes...

connected the conditioning instead of the string, everything is fine now.

willow tendon Mar 5, 2024, 5:31 PM

#

main/4.0 version is now available for testing purposes https://github.com/skunkworxdark/metadata-linked-nodes/tree/4.0-test

willow tendon Mar 6, 2024, 9:47 PM

#

update to 4.0-test branch for some type changes to main + some addition update for typing

willow tendon Mar 11, 2024, 2:03 PM

#

update to 4.0-test branch for some more type changes to main

willow tendon Mar 14, 2024, 5:43 PM

#

further update to 4.0-test branch for some more type changes to main

willow tendon Apr 2, 2024, 2:13 PM

#

2nd Apr 2024 - 4.0 Update
The main branch has been updated to be InvokeAi 4.0 compatible and the 4.0-test branch removed. Run the update.bat/update.sh in the node folder to get the latest version.

If you require the 3.7 or older compatible version then there are tags on the repo at the appropriate points that can be used. See the readme in the repo for further details.

#

Metadata-Linked & Metadata From Image Nodes [4.0+]

willow tendon May 9, 2024, 10:05 AM

#

9th May 24

Updated to use invocation_api - no functional changes

willow tendon Jun 28, 2024, 7:31 AM

#

28 Jun 24

Minor fix for DenoiseLatentsInvocation import path introduced with Invoke 4.2.5
New github tag invoke-4.2.4 to reference Invoke 4.2.4 compatible version

naive dust Jul 2, 2024, 12:13 AM

#

Trying to recreate my 1.5 upscale method with these for XL and I'm getting an error with Metadata to SDXL Model.

[2024-07-02 01:08:02,375]::[InvokeAI]::ERROR --> Traceback (most recent call last):
  File "C:\StableDiffusion\invokeai\.venv\lib\site-packages\invokeai\app\services\session_processor\session_processor_default.py", line 129, in run_node
    output = invocation.invoke_internal(context=context, services=self._services)
  File "C:\StableDiffusion\invokeai\.venv\lib\site-packages\invokeai\app\invocations\baseinvocation.py", line 289, in invoke_internal
    output = self.invoke(context)
  File "C:\StableDiffusion\invokeai\nodes\metadata-linked-nodes\metadata_linked.py", line 819, in invoke
    model = get_model(model_key, context)
  File "C:\StableDiffusion\invokeai\nodes\metadata-linked-nodes\metadata_linked.py", line 215, in get_model
    raise Exception(f"Unknown model: {model_key}")
Exception: Unknown model:```

willow tendon Jul 2, 2024, 7:51 AM

#

could you share a workflow and image that causes this error

willow tendon Jul 2, 2024, 8:01 AM

#

naive dust Trying to recreate my 1.5 upscale method with these for XL and I'm getting an er...

is it possible that the image you are extracting the metadata from has a LoRA that doesn't exist in invoke anymore?

naive dust Jul 2, 2024, 2:17 PM

#

willow tendon is it possible that the image you are extracting the metadata from has a LoRA th...

Images I was testing with were only just generated in t2i so LoRAs are all there. Should be able to get you the workflow/image to you soon

willow tendon Jul 2, 2024, 2:23 PM

#

naive dust Images I was testing with were only just generated in t2i so LoRAs are all there...

the error you supplied indicates it couldn't find the LoRA from the metadata. So instead used a LoRA called "" instead. I should just skip over a lora if it is not found and not try use one called "" (doh). I will fix this soon but I would like to confirm that this is the error you got or if something else happened. So the image with the metadata intact would be useful to track this down.

naive dust Jul 2, 2024, 2:24 PM

#

I'm on it now, no problem.

#

📎 Upscale.json

willow tendon Jul 2, 2024, 3:05 PM

#

naive dust I'm on it now, no problem.

Thank you. I have a bug in my extracting LoRA from metadata code. working on fixing it now.

willow tendon Jul 2, 2024, 4:05 PM

#

02 Jul 24 - There is a new version with a fix to the model identification. It will now try load the model via the model key in the metadata then if that fails it will try using the model name. New versions of metadata_to_loras and metadata_to_sdlx_loras however this change will also effects any model loading done by the nodes

#

@naive dust new version in the repo should fix your issues

naive dust Jul 2, 2024, 5:16 PM

#

willow tendon <@187852551293632512> new version in the repo should fix your issues

Great work. Thank you. I'll give it a try shortly.

naive dust Jul 2, 2024, 5:31 PM

#

willow tendon <@187852551293632512> new version in the repo should fix your issues

Working wonderfully. Thanks again. These are invaluable.

willow tendon Jul 2, 2024, 5:34 PM

#

naive dust Working wonderfully. Thanks again. These are invaluable.

especially now the gallery metadata search is on the horizon.

naive dust Jul 2, 2024, 5:37 PM

#

willow tendon especially now the gallery metadata search is on the horizon.

Oh, I haven't heard about that. Sounds extremely useful.

willow tendon Jul 2, 2024, 5:40 PM

#

it was added on the main branch a couple of days back so I guess it will be in a full release in the not too distant future.

naive dust Jul 2, 2024, 5:42 PM

#

That'll be great. It was really useful in that Breadboard image management app until it was abandoned.

willow tendon Sep 28, 2024, 4:44 PM

#

New Release available :

28 Sep 24 - invoke 5.0 Flux update. Added Flux Denoise + metadata node. (auto collects model, seed, width, height, steps, guidance, denoise_start, denoise_end and lora metadata.

willow tendon Jan 23, 2025, 8:44 PM

#

New Release available:

23 Jan 25

Added flux specific built in metadata names guideance, cfg_scale_start_step & cfg_scale_end_step
Added Automatic recording of flux specific metadata to FluxDenoise node - cfg_scal, cfg_scale_start_step & cfg_scale_end_step
new node Metadata To LoRA Collection - Creates a Lora Collection from metadata that can be fed into the Lora Collection Loader nodes.

Warning if no LoRA metadata is available then this will cause an error in the Lora Collection Load nodes. There is a PR in the main InvokeAI to fix this. https://github.com/invoke-ai/InvokeAI/pull/7579

GitHub

LoRA Collection Loader make optional LoRA Collection input by skunk...

Update the LoRA Collection Loader nodes to make the Lora Collection input Optional
Summary
LoRACollectionLoader, SDXLLoRACollectionLoader & FLUXLoRACollectionLoader changed to make the LoRA...

willow tendon Jan 29, 2025, 6:26 PM

#

29 Jan 25

added a custom label to the Metadata to LoRA Collection node. Now you can pull and store lora data under any metadata label name.

willow tendon Feb 27, 2025, 1:53 PM

#

Metadata-Linked (now part or core as of invoke 5.7)

#Metadata-Linked (now part or core as of invoke 5.7)