Instructions to use h94/IP-Adapter-FaceID with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use h94/IP-Adapter-FaceID with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("h94/IP-Adapter-FaceID", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps
- Draw Things
- DiffusionBee
The old `ip-adapter-faceid_sdxl_lora.safetensors` seems to be better than the new
I've been using FaceID-SDXL for a long time with the previous LoRA checkpoint (sha256: 2ade2b53dddff5e96139400f68525c38f286d0597db6f891ba08629a6ade5b20), I've been throw 5 hour of debug to understand why my performance had degrade a lot. Seems to be related to the new LoRA checkpoint (sha256: 4fcf93d6e8dc8dd18f5f9e51c8306f369486ed0aa0780ade9961308aff7f0d64).
Could you elaborate a little bit on this commit ?
Here is a before after comparaison with exactly the same comfy workflow:
Target:
New LoRA:
Previous LoRA:
Update:
Seems that the previous LoRA wasn't loading well (I have a lot of key error in the comfy console). And so it was not using the LoRA. I will investigate but seems that with my current config the LoRA improving anything and even degrading the performance. I will keep make update here if I figure out why ?