refactoring + standalone api

[graemeniedermayer]

This pull request enables api to work with in standalone mode (the auto-generated docs are still wrong).

---

Notes

This implements many suggestions/changes from @semjon00 . I'm not entirely sure how @semjon00 should be credited.

Some of the alterations from the draft #316.

• It keeps the output of the api the same. (The current outputs still feels poorly structured)
• it does break api inputs a little bit.
• remove gradio imports from api directories.
• because depthmap_api.py was so small it's contents are append to the end of depthmap.py

I have only tested it with core features. So things like 3d/video are unlikely to function.

---

How to use

The standalone api will be enabled with the options --api. The full start-up becomes python main.py --api.

sample api request (python 3.8)

import requests
from io import BytesIO
import base64
from PIL import Image
import json

with open("test_img.png", "rb") as image_file:
        img = base64.b64encode(image_file.read())

dics = {
  "input_images": [img],
  "generate_options":{
    "compute_device": "GPU",
    "model_type": "zoedepth_n (indoor)",
    "net_width": 512,
    "net_height": 512,
    "net_size_match": False,
    "boost": False,
    "invert_depth": False
  }
}

# this points to auto1111. this is usually the default local address.
url = 'http://127.0.0.1:7860/depth/generate'

#This is where the request is made
x = requests.post(url, json = dics)
print(x.text)
img = json.loads(x.text)['images'][0]
print(img)
#Image is decoded here. this might have changed in python 3.9
img_io = BytesIO(base64.b64decode(img))
im = Image.open(img_io)

#Image is
im.save('image1.png', 'PNG')

[graemeniedermayer]

should I update this?

[semjonsona]

Hi 😅
I got quite busy and could not really give much time to this. I am sorry.

My concern is breaking inputs - could you please try to do it so that there is no change? If API behaves exactly the same, but can be accessed without Gradio (and code is reasonable) - it would be a bliss and easy to merge.

[semjonsona]

Really would be nice to have it. People tell that API it is useful (for example aulerius)

[graemeniedermayer]

Things have been soo busy recently! I'll keep the API input the same.

[davidmartinrius]

Hi, how are you doing? This code is right, but if accepted maybe it will be deprecated soon as there is a project dedicated for the stable diffusion API: https://github.com/mix1009/sdwebuiapi

I am a contributor of that project and I am working in an endpoint to do the same task with depthmap script.

Also I added a new funcionality to generate videos with this project, without affecting the core.

Soon there will be a new pull request to this project and sdwebuiapi.

I added a new option RUN_MAKEVIDEO_API with parameters such as num of frames, fps, xyz movements, etc

Thank you!

David Martin Rius

[semjonsona]

Hello @davidmartinrius, nice to see you here! Honestly, I am even more busy with stuff nowadays, but I am looking forward for your MR, hope I can go through it quickly-enough 🤞🫠

[graemeniedermayer]

@davidmartinrius This sounds great! I was unaware of this repo thanks for linking it.

[davidmartinrius]

Hello! I created a new pull request in #378

I hope it fits to the project standard 😄

David Martin Rius