can't replicate img2img with api

[sould3mon]

im trying to replicate trough the api.

api-generated pngifo:
Steps: 36, Sampler: DPM++ 2S a Karras, CFG scale: 10.0, Seed: 1144868451, Size: 512x512, Model hash: d8722b4a4d, Denoising strength: 0.8, Clip skip: 2, ControlNet-0 Enabled: True, ControlNet-0 Module: depth_leres, ControlNet-0 Model: control_sd15_depth [fef5e48e], ControlNet-0 Weight: 1.0, ControlNet-0 Guidance Start: 0.0, ControlNet-0 Guidance End: 1.0, ControlNet-1 Enabled: True, ControlNet-1 Module: openpose, ControlNet-1 Model: control_sd15_openpose [fef5e48e], ControlNet-1 Weight: 1.0, ControlNet-1 Guidance Start: 0.0, ControlNet-1 Guidance End: 1.0

webui:
Steps: 36, Sampler: DPM++ 2S a Karras, CFG scale: 10, Seed: 1144868451, Size: 512x512, Model hash: d8722b4a4d, Denoising strength: 0.8, Clip skip: 2, Mask blur: 4, ControlNet-0 Enabled: True, ControlNet-0 Module: depth_leres, ControlNet-0 Model: control_sd15_depth [fef5e48e], ControlNet-0 Weight: 1, ControlNet-0 Guidance Start: 0, ControlNet-0 Guidance End: 1, ControlNet-1 Enabled: True, ControlNet-1 Module: openpose, ControlNet-1 Model: control_sd15_openpose [fef5e48e], ControlNet-1 Weight: 1, ControlNet-1 Guidance Start: 0, ControlNet-1 Guidance End: 1

the only diffrence is:
Mask blur: 4, and api contains float values

verified multiple times that the prompt is the same (taking api-generated pnginfo >take prompt to img2img

steps taken in webui:
add img2img file
add 2 images to controlnet
enable controlnet (default values)
set the settings as seen in png info

i just can't figure out where this mask_blur comes from in the webUI
tried some things with the api but api shows:
"mask_name": "string" ,
"mask_blur": 4,
and can't figure out what the webUI uses

api-code:

setunit ={
 "init_images":[inpimg],
 "denoising_strength": 0.8,
 "prompt": vprompt,
  "steps": 36,
  "batch_size":bsize,
  "restore_faces":False,
  "negative_prompt": f"{neg_promt_1}", #{neg_promt_2}{neg_promt_3}
  "cfg_scale": 10,
  "width": 512,
  "height": 512,
  "seed": seed ,
  "sampler_name": "DPM++ 2S a Karras",
  "seed_resize_from_h": 0,
  "seed_resize_from_w": 0,
  "resize_mode": 0,
  "mask_name": "" ,    #????
  "mask_blur": 4,
  "controlnet_units":[
     {
     "input_image": img ,
     "module": "depth_leres",
     "model": "control_sd15_depth [fef5e48e]",
     "weigth": 1,
     "guidance": 1,
     "guidance_start": 0,
     "guidance_end": 1,
     },
     {
     "input_image": img ,
     "module": "openpose",
     "model": "control_sd15_openpose [fef5e48e]",
     "weigth": 1,
     "guidance": 1,
     "guidance_start": 0,
     "guidance_end": 1,
     }
  ]
 }
 return setunit

[dbokser]

Did you ever get this figured out? Seems like you're looking for "mask" not "mask_name"?

[sould3mon]

it's been a bit but i copied some example maybe from the api doc's or a script as an example

                        image = Image.open(io.BytesIO(base64.b64decode(i.split(",",1)[0])))
                        png_payload = {
                            "image": "data:image/png;base64," + i
                        }

                        response2 = requests.post(url=f'{url}/sdapi/v1/png-info', json = png_payload)

                        pnginfo = PngImagePlugin.PngInfo()
                        pnginfo.add_text("parameters", response2.json().get("info"))
                        seedinfo = response2.json().get("info")

                        image.save(output_dir+"test_P"+f"{prompt_no}_{grab_seed(seedinfo)}_"+f"{counter:04}" +".png", pnginfo = pnginfo)

and then drop the result back into the webui png info tab and for the generated image the same either from the direct result at the bottem or dropping result image in the same tab.

but at this point it's bit low priority for me this field progresses so fast that we can wait couple weeks and eveything we do has changed
again.
so at some point ill just need to refactor the code or the whole method got obsolete and we get a new way of doing things.

although i do appreciate that the question still get read and thought about.

[matangrinberg]

Ah I understand. And how are you inputting the image? I cannot seem to get the init_image formatting to work to replicate this issue.

[sould3mon]

here it gets a bit messy why i need to refactor .
i used https://youtu.be/_xHC3bT5GBU technique : lightmap +controlnet in this script

written one function wich (misnamed for input image i only now realize why imp_img ????)

def imp_img(name):
    with open(name,"rb") as cimg:
        istr = base64.b64encode(cimg.read()).decode('utf-8')
    return istr

later on also added one to it's own package/file when working with PIL stuff thats already in python like when returned from stable diffusion

def get_imgstr(img):
    buf = io.BytesIO()
    img.save(buf,format="PNG")
    strimg=base64.b64encode(buf.getvalue()).decode('utf-8')
    return strimg

wich are fed into the dictornary in the first post :
"init_images":[inpimg], (uses imp_img file from disk)
"input_image": img , (uses get_imgstr loopback img returened from stable diffusion)

i should rewrite it all when i got time for it i hacked this in a weekend session so it has become quite messy.

[kingpreyansh]

were you able to find a solution for this?

[asdfzxh8]

etc wrg, no such thing as nx or know or better or etc, cepuxuax, asx any nmw s perfx