bedovyy
Add Support for V4.5 Inpaint with Strength and V4.5 Vibe
Have been using the node and it is more efficient compared to Novelai's frontend. Just want these 2 functions:
-
The current inpaint node does not have strength parameter compared to Novelai's frontend. Could it be added? Also is the V4.5 inpaint supported?
-
V4.5 vibe transfer just came out. If it can be added it would be extra helpful!
I added those features and character prompts (that was already working but commented out for some reason).
I don't want to make a fork or anything so I'll just upload the updated module.
The vibe transfer node needs to be put after model option node to detect v4+ mode. Also, don't use the mask convert node, it's already converted in the other nodes and doing it twice will make strange results.
Also, vibe encodings for v4+ use 2 anlas. The encodings are saved and reused, but they are unique per each different "information extracted" parameter value and mask.
I keep getting error 500 with the vibe transfer stuff @tamixy . Is there anything outside of VibeTransfer option and load image needed for it to work?
If you're getting the error on generate and not the vibe transfer node, it's probably because you didn't pass the model option to the vibe transfer node as I said. If the model option isn't passed, it will default to v3 vibe transfer which is incompatible.
Put the model option node first in the option transfer stack, then vibe transfer, and the rest.
If this is added properly to the extension, it might be better to put the version type as an input on the vibe transfer node instead of requiring the option. Or put the vibe extraction api requests inside the image generation node.
Also, another note, vibe encodings for v4+ use 2 anlas. The encodings are saved and reused, but they are unique per each different "information extracted" parameter value and mask.
~~It should have been passed properly. Is this how it's done? I have anlas and opus and it works completely fine without the Vibe stuff.~~
Nevermind, it needed a V4 prompt config node, I thought model selection was enough. Thanks.
Sorry, I was just about to reply, yeah, I didn't realize the option I was using to check for v4 was from V4 prompt config. I'll update the files, that's not intended.
Thanks! I am pretty sure to original guy would appreciate a git pull for the changes.
I give permission to diff check the files and add the changes, I just don't want to post code on github because then they block your login unless you add a phone number.
Oh my....I already thought the node is discontinued......Greatly appreciate the help. Cannot wait to test it out!
@tamixy Hi, yeah the new update works like a charm. Thanks again for the contribution!
A small bug I noticed:
The autosave_filename_prefix parameter in the Generate node, I notice by default it is "false,true" now. Did you want to change it to a toggle? Coz currently it is still the filename prefix and the autosave is still always active.
I'm not getting that bug, maybe try creating a new node, it sounds like something that could happen if you load an updated node on an old workflow.
I'm not getting that bug, maybe try creating a new node, it sounds like something that could happen if you load an updated node on an old workflow.
Add a new generate node and they parameter is just "NAI_autosave". I guess it is just not a toggle.
Anyway thanks!
Another question on one of the new node, the V4 base prompt node:
What does the node do? It looks just like a text box to me, the output is just text string. Is it the same just as text box or it has some specific use? The character caption can be connected to other NAID Option that is intuitive.
I have no idea, I didn't make it. Looking at the code it just transfers the input to the output untouched. Comments on the code suggest it used to do something but was removed.
Thanks for the reply. Ah then maybe just an abandoned one, no wonder why I did not notice it earlier....
@tamixy Hi, just want to follow up a bit for the update, which has been splendid. Just wonder: I saw there is a "normalize reference strength" button in Novelai's web but in the node I did not see such thing. Is it normalize them by default, or it need to be implemented seperately?
Dunno, there's nothing about it in the api docs. Maybe it adjusts the vibe strengths client-side by reading the vibe info values.
On another similar note, I've seen that the image gen api has a couple of controlnet parameters, does anyone know anything about that? It's separate from director tools, is it a removed old feature, or possibly something upcoming?
Yeah I saw the controlnet thing as well. I think they used to have something like that:
https://blog.novelai.net/novelai-controlnet-tools-upscaling-new-image-generation-ui-8e3c2b174acb
But it is no longer there anymore I guess
@tamixy Thanks so much for your amazing work, the code runs smoothly!
I just have one question: I’m not sure if this is a bug, but when generating two or more characters, the local outputs look different from the online generations, even though I imported the exact same metadata. Or maybe I linked the nodes incorrectly?
For reference, my character node setup is:
v4 character caption 1 → v4 character caption 2 → network option → model option → Generate
Thanks in advance for the help!
@FoxZzz I don't know, if you have character coordinates on the website, that needs to be turned on in the V4 prompt config node.
Not sure if this is the right thread for it, but do you have any plans on adding support for Character Reference too? Would be really helpful! https://blog.novelai.net/preview-release-novelai-character-reference-for-consistent-characters-dc882be996c7
I’m not too familiar with GitHub issues, so I’m not sure if @tamixy noticed your comment. It might be worth tagging directly here.
I opened a separate ticket about Character Reference. There are a few requirements such as the resolution size of each image. I'm not really a coder, but I was able to gather this info from looking at API calls when generating on official NovelAI. It may not be straightforward to implement because it asks for an image of a larger resolution than what you're generating (if you're at the standard sizes).
Updated extension with character reference.
ComfyUI_NAIDGenerator_char_ref_update.zip
Image resizing functionality is taken from KJnodes.
Image output on the character reference node is just for previewing, only the option is needed. It can be inserted at any position among other nai option nodes.
Updated extension with character reference.
ComfyUI_NAIDGenerator_char_ref_update.zip
Image resizing functionality is taken from KJnodes.
Image output on the character reference node is just for previewing, only the option is needed. It can be inserted at any position among other nai option nodes.
Downloading this and overwriting my install of the custom node with it broke it.
Updated extension with character reference. ComfyUI_NAIDGenerator_char_ref_update.zip Image resizing functionality is taken from KJnodes. Image output on the character reference node is just for previewing, only the option is needed. It can be inserted at any position among other nai option nodes.
Downloading this and overwriting my install of the custom node with it broke it.
What's the error?
Updated extension with character reference. ComfyUI_NAIDGenerator_char_ref_update.zip Image resizing functionality is taken from KJnodes. Image output on the character reference node is just for previewing, only the option is needed. It can be inserted at any position among other nai option nodes.
Downloading this and overwriting my install of the custom node with it broke it.
What's the error?
It's okay. It's my bad for not reading through the replies here. It was definitely the order my nodes were in. Rearranging them fixed it.