Instructions to use nphSi/Z-Image-Lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use nphSi/Z-Image-Lora with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Tongyi-MAI/Z-Image,Tongyi-MAI/Z-Image-Turbo", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("nphSi/Z-Image-Lora") prompt = "Alexandra Chando (vrtlAlexandraChando)" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps
- Draw Things
- DiffusionBee
Request / Suggest a Lora
You can request as much Loras you like but put each in an extra post here (not thread). As soon as your post gets 4 thumbs-up-votes i start baking!
- Currently accepted: Z-Image.
- Priority of Requests/Suggestions, Number of new Loras and update frequency is directly related to the amount of support i receive via ko-fi.com
- I may fulfil requests with less than 4 upvotes if i am personally interested. ko-fi.com can raise my interest!
- Well known Characters only. No Influencers or TikTok/OnlyFans/Adult models or Pornstars etc. This belongs to B-Sides and requires support via ko-fi.com (see above).
- About requests for a specific Era or themed Lora:
The most important thing about making a good working Lora is absolute highest quality and diversity in the dataset. Being stuck in a tight time frame with analogue magazine scans
makes it impossible. Such a Lora would be highly inflexible and from bad overall quality. See Ornella Muti or Cybill Shepherd Loras for example. They are made exclusively with old
analogue photo scans. Another example is Stephanie Meissner, made out of a few 720p Youtube clips. My goal is always to reflect people when they where at their prime time while
matching the image we have memorised them as close as possible.
The major criteria for Loras in the main list is broad mainstream popularity and availability of good quality and diverse images on free and public accessible sites.
No Supermodels, "Miss Whatever" or anyone whos main business is being photographed. Exception are (Ex) Supermodels who turned actor etc.
Before requesting/suggesting a Lora please do yourself a basic check for available images for a dataset. Check listal.com and wikifeet.com . If you do not find some images there a Lora is probably not possible at all.
If you have already collected images usable for a dataset please provide a download url.I will silently ignore all entries from people who clearly did not read this post.
🧠 = Training
➕ = Finished
😔 = On hold (few/bad images)
Myoui Mina from Twice
Jodie Whittaker
Mandip Gill
Millie Gibson
Freema Agyeman
Pearl Mackie
Bianca Belair (WWE)
Emily Armstrong (Linkin Park)
Laura Müller
Jella Haase
OMG thx so much for completing my request <3
Humbly I ask,
Amanda Holden
Daisy Lowe
Myleene Klass
Your work is so great, can you plz post your Kofi link more than happy to contribute!
Note
Currently training at RunPod is nearly impossible. Updates will be much slower. Please consider this if you’re planning to support me with the intention to request B-Sides Loras.
Same for vast.ai and others , it's probably because of crypto mining for Pearl: e.g. https://www.binance.com/en/square/post/328328689035569
Hey @nphSi , just bought you a "Finger" on Ko-Fi, whatever the hell that means.
Thanks for all your amazing work!