miluELK

/

Taiyi-sd-pokemon-LoRA-zh-512-v2

like 1

I'm so sorry I haven't check my community recently.I've just use your models to generate 2 images in the same parameters,your work is very effective.The image of LoRA looks better.Perhaps there is a error in your generation code,here's my code:

import torch
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained("deepgoyal19/mysd1")
pipeline.to("cuda")
generator = torch.Generator("cuda").manual_seed(2000)
prompt = "cute puppy"
image = pipeline(prompt, generator=generator, num_inference_steps=50, guidance_scale=7.5).images[0]
image

sd1

from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained("deepgoyal19/mysd1",torch_dtype=torch.float16)
model_path = "deepgoyal19/lora15"
pipe.unet.load_attn_procs(model_path)
pipe.to("cuda")
prompt = "cute puppy"
generator = torch.Generator("cuda").manual_seed(2000)
image = pipe(prompt, generator=generator, num_inference_steps=50, guidance_scale=7.5).images[0]
image

lora

Thankyou for your reply.

As you have seen my LORA generated good results but it this is not happening on Hugging face inference api.(https://huggingface.co/deepgoyal19/lora15)

Do you know why?

This situation is normal,you will find more imformation in https://api-inference.huggingface.co/models/deepgoyal19/lora15,the JSON shows that the argument "id: deepgoyal19/lora15" and "base_model: deepgoyal19/mysd1",if you use this model id in the pipeline like the following code:

from diffusers import DiffusionPipeline
pipeline = DiffusionPipeline.from_pretrained("deepgoyal19/lora15")

you will find this error:

 HTTPError: 404 Client Error: Not Found for url: 
Entry Not Found for url: https://huggingface.co/deepgoyal19/lora15/resolve/main/model_index.json.

because there is no model_index.json or other files for pipeline,so it can't find such files.
I guess in order to avoid this error while using API,the programmer of HuggingFace just use the base_model id for pipeline,the API is using model "deepgoyal19/mysd1".

In order to solve this problem, I have two idea:

• 1. maybe add the model "deepgoyal19/mysd1" files into "deepgoyal19/lora15" repository will solve this problem ,I'm not sure ,you should backups these two repositories if you want to take a try.
• 2. the HF API is unreliable,I think they use "Gradio" and "FastApi" lib to bulid it.Maybe custom a API in your Space is a better choice,you could find a introduction in this page Quickstart

I hope it will be help.

I tried your idea no.1 and it's not working.
I know that I can make my custom space but HF API is really fast so I did want to do this with that.

I am still not able to figure out why only my model is not working. I can see everyone's LORA model working on hub.

Thankyou again!