-
Notifications
You must be signed in to change notification settings - Fork 1
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Lora weights are incorrect #5
Comments
Civitai doesn't recognize that field as it belongs to SwarmUI most probably |
I thought the application responsible for converting the meradata to civitai standards. So this would be a bug, if the weights are not converted? |
It only converts the parameters that are visible in the site like the prompt, negative prompt, sampler, resources used, etc. Every other info of the image from Swarm is added to the metadata as information in case someone wants to copy the raw metadata from civitai |
But these parameters are visible inside swarmui. They are also reusable, but not for civitai. |
this extension was made just for civitai to be able to read and parse the metadata automatically so it can set things like the prompt you used, the resources like loras used to make the image, etc. |
Yes, but the loras used are not converted with the correct weights. I know civitai uses A1111 standard format, so I thought the extension is converting swarmui format to a1111 and in this case the weights are missing. Right? |
Civitai added Swarm metadata parsing a while back so I'm not updating the extension. This extension only formats the swarm metadata to A1111 and it does convert some fields to be readable by civitai, it probably doesn't read the field called |
That's weird, civitai support wrote me a mail that they only parse a1111 information. Also if I upload a image without extension they can not recognise any model or lora. |
Yea, they didn't modify the parser to support model recognition in Swarm metadata since it was added recently. Also as far as I know they support A1111, ComfyUI and SwarmUI metadata |
The support stated they only support a1111, not swarmui metadata... Currently, our metadata parser uses the standard A1111 format, but we've received a few feature requests related to this topic on our Feedback page. If there are any specific suggestions, you're more than welcome to chime in with a feature request of your own, or you can upvote someone else's idea for consideration in future updates to the site.) ... So there is no way they use the correct metadata weights? |
It does support Swarm because I've tested it, you can also test yourself. As for the weights the key name that is used in A1111 might not be |
I did try it yesterday, if I upload a picture crated from swarm, the lora and model is missing and there is a suggestion to use civitai helper or something. How do you did the generation to recognise the swarm metadata? |
Metadata is not only the resources, it includes prompt, seed, schedulers, etc which is automatically detected. As I said before civitai has to modify their parser to detect the resources used because Swarm implemented that but they didn't update their tool yet |
Crazy, I did wrote the swarmui creator and he didn't want to look into that "they are responsible" he said. And civitai said they only do a1111 ... Maybe it is a dead end. But that's sad, so nobody can reuse the data to recreate from the metadata shown there. Can you tell me what swarm integrated? The compatibility's to civitai? |
Swarm added the hashes of the resources you use so that crossposting can be done but Civitai didn't implement that to their metadata parser yet |
All Lora weights are interpreted as "1" inside civiai upload. Like in this example the weights are 0.33...0.2 and so on, but after uploading to civitai all weights are shown as "1".
Model hash: 86d99a344203, Model: illustrious/DaSiWa-Illustrious-XL-0-4-dynMa.safetensors, Lora hashes: "748cmSDXL: a4f81ec6c80b, dasiwa-ill-colorful-detail-style-1561-39: 3acf9bb397af, GBF_Illustrious: e7450d76b275, GENESIS_MK0.2: 8e936ac710d4, MGE_V5.3: 41e4f5016d0d, ponyv5_noobV065S_1_adamW-000013: 3c03b2d886f9, USNR STYLE_XL_lokr: 74988f37b9f2", aspectratio: 3:4, refinercontrolpercentage: 0.2, refinermethod: PostApply, refinerupscale: 2, refinerupscalemethod: model-Foolhardy_Remacri_x4.pth, automaticvae: True, loraweights: "0.33,0.3,0.3,0.3,0.33,0.22,0.22"
The text was updated successfully, but these errors were encountered: