Voyage: Let's explore unexplored sea of AI Art

What is Voyage?

Voyage is basically a text to image model developed by Muhammadreza Haghiri and it is based on weights from Stable Diffusion version 2.0 and Midjourney version 4. This model helps creative people turn their ideas to artwork for free (in any sense of the word free).

How to use voyage with diffusers lib

Installing needed libraries

!pip install --upgrade git+https://github.com/huggingface/diffusers.git transformers scipy ftfy accelerate

Importing required libraries, functions and classes

These following libraries, functions and classes used by me in order to test the model. Feel free to add more of your need or remove unnecessary ones!

from diffusers import StableDiffusionPipeline, EulerDiscreteScheduler, DiffusionPipeline, DPMSolverMultistepScheduler
import torch

Setting up the model and scheduler

In order to get results like what I got, you have to set euler scheduler up. This is how you can get it:

model_id = "openvoyage/voyage-v1"

scheduler = EulerDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, torch_dtype=torch.float16)
pipe = pipe.to("cuda")

but if you need DPMS scheduler, you can use this line as well:

scheduler = DPMSolverMultistepScheduler.from_pretrained(model_id, subfolder="scheduler")

Inference

prompt = "ultra realistic illustration of a young beautiful woman, intricate, elegant, sharp focus, smooth edges"
negative_prompt = ""
prompt = f'<voyage> style {prompt}'
image = pipe(prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=25, width=512, height=512, guidance_scale=10).images[0]

and in order to save your images, you can use image.save() method and have it in PNG format.

Colab notebook

This github repository belongs to the responsible colab notebook of the model.

Samples

figure 1 figure 2 figure 3 figure 4

Downloads last month
3
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.