Api call

[saaldigital-cw]

How can i use the refiner within an api call. I tried that but did non work?

payload["alwayson_scripts"]= {
"refiner": {
'args': [
{
'enable': True,
'model_name': item["StableDiffusionParams"]["Redefiner"]["Model"],
'steps': item["StableDiffusionParams"]["Redefiner"]["Steps"]
}
]
}
}

[altoiddealer]

Also interested in this

[saaldigital-cw]

I found a way..you have to add something like this:

payload["alwayson_scripts"]= {
"refiner": {
'args': [
True, modelnameWithId,NumberSteps
]
}
}

[markheaps]

I did this, but not sure how I know if it's using the refiner or not. I didn't get an error, but at the same time didn't see any indication that it was used.

payload = {
"prompt": prompt,
"negative_prompt": negative_prompt,
"batch_size": batch_size,
"steps": base_steps,
"width": width,
"height": height,
"sampler_index": "Euler a",
"seed": seed,
"refiner": {
'args': [True, "sd_xl_refiner_1.0_0.9vae.safetensors [8d0ce6c016]",refine_steps]
}
}

INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 10 images in a total of 1 batches.
100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [01:49<00:00, 3.14s/it]

whereas when I run it in the webui, the first thing its says is its loading the refiner
Loading refiner...

100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:22<00:00, 1.52it/s]
Total progress: 175it [15:25, 5.29s/it]
Total progress: 175it [15:25, 1.11s/it]

[altoiddealer]

It would be great if the refiner added something to the metadata… been meaning to put in an enhancement suggestion in Refiner repo (probably one there already)

[saaldigital-cw]

-You can check it if you use the same seed+see if the image look different
-In the console i see that he will load the redefiner model and make a small pause before processing the last refine_steps steps

[markheaps]

Yea best I can tell its not doing anything. No pause, the picture seems to be the same.

But yes, I notice the pause in when running from the GUI so I know what your talking about.

[altoiddealer]

So I did make an enhancement suggestion in Refiner repo, to add metadata to the output images (model + steps).

they did implement this yesterday, so after you update the extension, you should be able to easily tell if the refiner was used or not by just looking at the metadata.

[altoiddealer]

I did this, but not sure how I know if it's using the refiner or not. I didn't get an error, but at the same time didn't see any indication that it was used.

payload = { "prompt": prompt, "negative_prompt": negative_prompt, "batch_size": batch_size, "steps": base_steps, "width": width, "height": height, "sampler_index": "Euler a", "seed": seed, "refiner": { 'args': [True, "sd_xl_refiner_1.0_0.9vae.safetensors [8d0ce6c016]",refine_steps] } }

INFO:sd_dynamic_prompts.dynamic_prompting:Prompt matrix will create 10 images in a total of 1 batches. 100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [01:49<00:00, 3.14s/it]

whereas when I run it in the webui, the first thing its says is its loading the refiner Loading refiner...

100%|██████████████████████████████████████████████████████████████████████████████████| 35/35 [00:22<00:00, 1.52it/s] Total progress: 175it [15:25, 5.29s/it] Total progress: 175it [15:25, 1.11s/it]

You're close. I was able to figure out what you were missing, and also another thing that could be preventing it from working.

1. Make sure you are using latest version of Refiner extension
2. Make sure that the hash of the weights is visible in the checkpoints dropdown of your A1111 UI for your refiner model (don't worry about the dropdown in Refiner tab). By default, it will not have the weights there until you load it as a checkpoint. (Make sure "[8d0ce6c016]" is after the refiner model name in A1111 UI).
3. If you have the refiner model in the models/Stable-diffusion directory then you have the correct arg for the model. However, if you use a subfolder like I do, then you need to update this arg to include the subdirectory followed by a double backslash.
   In my case I have it in a subdirectory "sdxl" so my arg is:
   "sdxl\\sd_xl_refiner_1.0_0.9vae.safetensors [8d0ce6c016]"
4. Replace "refine_steps" with the value for the percentage of refiner steps. Default and recommended is 20.

With all that said here is what is working for me

        'alwayson_scripts': {
            'refiner': {
            'args': [
                True,
                "sdxl\\sd_xl_refiner_1.0_0.9vae.safetensors [8d0ce6c016]",
                20
            ]
            }
            }

As with using it in the A1111 UI, the cmd window will indeed show "loading Refiner" on first gen, and you will be able to see the metadata in the image confirming this.

[markheaps]

Can you point me in the right diretion of where to look for 2) "Make sure that the hash of the weights is visible in the checkpoints dropdown of your A1111 UI".

I've completed 1, 3 and 4. After generation the picutre metadata only shows the mbb model used in in image browsing,
Model hash: 8b322c8033, Model: mbbxlAlpha_v01

whereas one completed by the GUI shows boththe mbb and refiner.
Model hash: 8b322c8033, Model: mbbxlAlpha_v01, Refiner model: sd_xl_refiner_1.0_0.9vae, Refiner steps: 4,

Here is my latest code snips

payload = {
"prompt": prompt,
"negative_prompt": negative_prompt,
"batch_size": batch_size,
"steps": base_steps,
"width": width,
"height": height,
"sampler_index": "Euler a",
"seed": seed,
"refiner": {
'args': [True, "sd_xl_refiner_1.0_0.9vae.safetensors [8d0ce6c016]",refine_percent]
}
}

option_payload = {
    "sd_model_checkpoint": "mbbxlAlpha_v01.safetensors [8b322c8033]"
}

response = requests.post(url=f'{url}/sdapi/v1/options', json=option_payload)

response = requests.post(url=f'{url}/sdapi/v1/txt2img', json=payload)

[altoiddealer]

Your code snipet reads “refine_percent” where it should read “20”

from your screenshot, the hash is visible in the UI “[8d0ce6c016]”. If you never loaded the model, it wouldn’t be displayed there and actually wouldn’t load it via api

[markheaps]

refine_percent is just a variable that I have set to 20. I can try hard coding it but don't expect any differance.

[markheaps]

Should "refiner" show up when I do:

http://127.0.0.1:7860/sdapi/v1/options

I don't see it if so.

[markheaps]

Ugg. Ok, figured it out. I looked at your example before and didn't think I needed 'alwayson_scripts': { wrapper. Once I added that it worked. If you don't wrap it in that and just have it directly as was my sample code, it doesn't work.

All good now.