Many people have a query in their mind Can we run Dall-e run locally? The answer is Yes. We have a new model called OpenDalle which can generate image art like what we can inOpenAI’s Dalle but using Stable Diffusion.
Not, only this, there are multiple issues related to the policies and restrictions of Dalle which sometimes problematic and can’t get the overall advantage of the black box model. But, using this model as open source you can get rid of such issues, widen your image generation leverage the power of Dalle into it.
We are showing this illustration on Automatic1111, but you can use this model in other WebUi or applications like ComfyUI, and Fooocus as well.
Install and run OpenDalle locally:
1. Go to the Hugging Face platform by using the link provided below:
https://huggingface.co/dataautogpt3/OpenDalleV1.1
2. Click on “File and Versions”.
3. You will be moved to the next page. Just scroll a little downward and search for “OpenDalle-Safetensors” which will be around 6GB file. Click on the download button to download it.
4. Now, save the downloaded model into your model’s folder. Restart the Automatic1111 to effect.
5. Here, after loading the OpenDalle model, you will see a new model in Stable Diffusion Checkpoint on Automatic1111.
Use the following settings:
-CFG Scale: CFG scale of 8 to 7
-Steps: 60 to 70 steps for more detail, 35 steps for faster results.
-Sampler: DPM2
-Scheduler: Normal or Karras
6. Now, let’s try with different prompts:
portrait of a black man, standing in a disco, GTA 5 concept art style
The results are really great. Observe the background with some disco lighting effects. Quite good. Now, let’s try for GTA5 girl.
portrait of a 20 years girl, wearing jeans, a jacket holding on her hand, standing on the side of the road, Gta5 concept style art
Run on Google Colab Using Diffusers Library:
from diffusers import AutoPipelineForText2Image
2. Import torch
import torch
3. Now, pass “dataautogpt3/OpenDalleV1.1” model as argument like this:
pipeline = AutoPipelineForText2Image.from_pretrained(‘dataautogpt3/OpenDalleV1.1’, torch_dtype=torch.float16).to(‘cuda’)
4. Input your prompts as an argument into a created object:
image = pipeline(‘black fluffy gorgeous dangerous cat animal creature, large orange eyes, big fluffy ears, piercing gaze, full moon, dark ambiance, best quality, extremely detailed’).images[0]