PhantomIXRowStyle - Preserving a Legacy Model
Seen it requested in several communities, the creator nuked it and all information associated with it but this model produces amazing results.
This is not my work and I don't plan to support this, just trying to preserve this amazing model for the community.
I don't have a safetensors version and won't be creating one.
Update: added safetensors version
(will keep the ckpt just for legacy reasons, they are both pruned)
Looking forward to seeing what people create with it.
Prompting / Info
Doing NSFW and SFW are equally possible and doesn't require much effort.
This model likes to do really detailed muscle line art so it helps to negative it out if you don't want it.
CLIP SKIP 1 or 2 seems to produce equally interesting results.
I don't show it - but this model works wonders in combination with LORAs and Hypernetworks.
This model is one of the best (IMO) for producing simplified color palettes without needing to go into crazy color tone prompting,
I love combining it with other styles to get a nice simplified palette and simple line work.
It has really creative lighting passes that are not overly complex.
Acknowledgments / Credits
I think the model author was https://huggingface.co/Cosk but the Cached Page got 404'd
Credit to the original artist PhantomIXRow
I never knew about them until I found this model and think their art is really fantastic.
Description
FAQ
Comments (13)
Can someone confirm if it's safe? Original owner deletes everything, OP refuses to make a Safetensors version, red flags make me want to err on the side of caution on this one.
Edit: It's cool now.
Then just DL it and run it through the scanner yourself.
Not sure about the safety of this model since they won't make a safetensors version, but the reason the original was taken down was that the person who made it got personally asked by the artist to take it down and they were chill and did that
Both HunggingFace and Civitai scan CKPT's for Pickle imports (You can see the validation status under the download button)
If someone is willing to create a safetensors - I'll upload it.
Use this one . it can direct DL - save to drive. there is one option to convert into safetensor https://colab.research.google.com/drive/1aQ5nXTfLWHhZi7GOfXteLKg5OT1X-aBZ
@juvesatriani My upload speed is awful.. I'll spend some free time looking at a local option.
@DreamExplorer I appreciate you, no personal offense by the post, i'd just rather not have my PC eat sh*t because I wanted to make something pretty.
@MrKissYourDad
safetensor uploaded now. Enjoy.
Converting to safetensor was super easy, took me about 2 minutes. Just my internet speed is terrible.
I tested a few prompts to make sure the images generated and they came out identical.
@DreamExplorer Thank you, love you!
@DreamExplorer actually you dont need to upload or DL anything , you can use your link from here ( DL link from your ckpt version) and let colab DL and converting yours into safetensor on the fly . And if needed you can save to drive or upload into huggingspace .
As far I`ve been tested , The colab I linked , not used our bandwith speed or data at all . That the main reason that colab created . So its handy to move from several online space into another . But it seem you`ve already upload the safetensor version . :)
Do you happen to know what model this was trained on? That's usually needed to extract a Lora from a model
Short Answer:
No Idea..
Long Answer:
Training Dataset:
https://huggingface.co/datasets/Cosk/misc-datasets/blob/main/styles/styles_notes.txt
The model is SD-v1 based..
CLIP has incorrect positions - Makes me think A3.. but not necessarily true.
They also used A3 for training other models (CSR)
https://huggingface.co/Cosk/sketchstyle-cutesexyrobutts
I thought there was a tool out there that could compare model weights but I can't remember the name of it.. honestly.. it could be anything that is SD-v1 based.. that's all I know..
I wish I saved the cached page.. it's gone now too.
Thanks for safetensors now I will download :)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.