API IMG2IMG

[craggy citrus]

Hello guys, I wanted to do this more detailed but my PC just died.

I was working on a small gradio app, using comfyui custom workflows as presets API from swarm t2i works perfectly but i2i didn't work

Using a swarm image base 64 input node, I had to manually copy paste the omg base 64 code 1m characters I think. the workflow worked
But in the swarm ui it wouldn't let me upload an image.

If I use an image input it works in the comfyui backend and in the swarm but it wasn't working in the gradio

So my question is there an example of a custom workflow, saved into a preset and used as an API call I can use to understand how to build the gradio app?

I've asked around to a few friends in the forum they helped a lot but I'm still stuck. If really appreciate an example on how to do this if someone had already done this please help.

[delicate skiff]

You should post the errors and the workflow.

The API calls would also help see what you are doing. Someone may be able to see the issue.

[craggy citrus]

so this is the basic comfyui workflow i staretd with somkething very simple so i can move onto more compelx workflows.

my problem is how im feeding the image to the workflow, base64 load image? TO USE IT VIA api PRESET

[craggy citrus]

Invalid operation: ComfyUI execution error: Incorrect padding

I GET THAT ERRROR USING SWARMINPUTIMAGE node

works in simple gen tab

[craggy citrus]

thats from the swamr log trying to use init image in generate

[craggy citrus]

using this my error is
}
2024-07-24 13:26:04,003 - INFO - API Response: {
"error": "Invalid operation: ComfyUI execution error: Incorrect padding"
}

using this
raise ValueError(f"API Error: {response_data['error']}")
ValueError: API Error: Invalid operation: ComfyUI execution error: cannot identify image file <_io.BytesIO object at 0x000001EB1B5D67A0>

using that i get a 748k character code that nukes my pc

[craggy citrus]

thats my gradio demo

2024-07-24 14:05:50,650 - INFO - Image2Image mode activated
2024-07-24 14:06:00,051 - ERROR - An error occurred: cannot identify image file <_io.BytesIO object at 0x0000019F71625080>
2024-07-24 14:06:24,498 - INFO - Image2Image mode activated
2024-07-24 14:06:24,734 - ERROR - API Error: Invalid operation: ComfyUI execution error: Incorrect padding
2024-07-24 14:06:35,470 - INFO - Image2Image mode activated
2024-07-24 14:06:42,877 - ERROR - An error occurred: cannot identify image file <_io.BytesIO object at 0x0000019F71624D10>

same errrors than in swamr

[craggy citrus]

I know it's probably something very dumb but idk how to fix this

[woeful frost]

oh, you didn't load the workflow here -- you just selected it from the dropdown, which only works when you have intercompatible params. When you have custom params you have to actually load it, to load the params

when calling from API, you'll want to look at how the Simple tab calls it

atm it's a bit hacky, the UI tracks the comfy workflow data and sends it up to the swarm server to reprocess

(ie the swarm server itself atm does not properly process the workflow on its own atm)

(this needs fixing, it is very silly that it works this way)

[craggy citrus]

Hello sir okay thank you so much I'll check the simple tab and how it calls the workflow I think I made it work once there

https://github.com/fidecastro/ComfyUI-serverless

I've been looking at this to interact with comfyui API directly

Doesn't know if it could be a solution? Since it wraps the whole workflow inside it . Maybe even allowing for text and image to video

[woeful frost]

you can use Swarm's APi and send workflows arbitrarily, or you can use Comfy's API through swarm (/ComfyBackendDirect routing), or you can use Comfy's API directly. That particular github repo seems to be a slightly questionable python launch wrapper for comfy

[craggy citrus]

Thanks I'll try this sir, thanks again for your help

[craggy citrus]

If you have any suggestions for sending images or video I'd gladly appreciate it 😅

[woeful frost]

images: base64

video, idk

[delicate skiff]

@craggy citrus This is the key.

[craggy citrus]

Okay thanks guys I won't sleep untill I fix this with this new help

Thank you so much

[delicate skiff]

@woeful frost uh, what am I doing wrong here I cant send init image to comfy workflow in generate tab either.

Create simple img2img workflow use SwarmLoadImageB6, save workflow, go to generate tab choose init image and toggle Custom Workflow with the one I just saved.

It errors because no image is loaded. Hitting F12 init image is included with my base64 in the call.

2024-07-30 21:45:05.030 [Debug] Refused to generate image for local: Invalid operation: ComfyUI execution error: cannot identify image file <_io.BytesIO object at 0x7ff8adc8ea20>

Does this not work the way I think it does? This is the same issue Onyx was having.

The workflow I used

[woeful frost]

Create simple img2img workflow use SwarmLoadImageB6, save workflow, go to generate tab choose init image and toggle Custom Workflow with the one I just saved.
This, again, is not the way to load comfy workflows on the generate tab

use the button

i needa hide that selector deeper down the Very Advanced Usages Only hole

[delicate skiff]

My thought process was whatever this call was from Swarm I could replicate how it calls the Comfy workflow and how it inserted the init image into the workflow so I could just use that as the api call in the code. i was trying to work in Swarm API not comfy but also still send an init image to the comfy workflow via the api.

Why does sending initimage param and comfyuicustomworkflow: test 2 not just load that image into the base 64 node and run the workflow?

[delicate skiff]

clicked send to generate tab chose image and clicked generate 2024-07-30 23:07:59.912 [Info] User local requested 1 image with model 'starlightXL.safetensors'... 2024-07-30 23:07:59.912 [Debug] [BackendHandler] Backend request #25 for model starlightXL.safetensors, maxWait=7.00:00:00. 2024-07-30 23:07:59.912 [Debug] [BackendHandler] Backend request #25 found correct model on #0 2024-07-30 23:07:59.912 [Debug] [BackendHandler] Backend request #25 finished. 2024-07-30 23:08:00.168 [Debug] Refused to generate image for local: Invalid operation: Collection was modified; enumeration operation may not execute.

[woeful frost]

this workflow has a manual parameter (SwarmInputCheckpoint) and therefore no parameters are autoclaimed, ie you do not have an Init Image input at all

[woeful frost]

fixed

[pine steppe]

hey @delicate skiff have you managed to make it work ? I cant find examples for img2img and Swarm API. Or at this time is it just better to use ComfyUI itself?
Im looking to make a script to generate images based on folder of init images and array of prompts to test things out with Flux.

[woeful frost]

it's very easy to use the swarm API for basic things, kaleb's only made a mess from the custom comfy workflow.
API docs @ https://github.com/mcmonkeyprojects/SwarmUI/blob/master/docs/API.md -- Swarm uses its own API, so you can just open your browser console and do a gen on the Generate tab, and look at the network traffic to see exactly how the api call works

[pine steppe]

Thanks @woeful frost only /Text2Image is mentioned here as route thats why i was wondering.

so calling /Imag2Image should work too

[woeful frost]

nono

look at the UI at all first before the API lol

"img2img" as some old UIs (auto) have called it, in swarm is just Init Image, a parameter on the same tab as everything else

[pine steppe]

I see thank you that makes sense

[woeful frost]

do a gen with an init image, and look at the network traffic, and that'll show you exactly how to do it as an api call

[pine steppe]

Yes will investigate it. Ive played with A1111 API but not the Comfy one yet and nor with Swarm. I think i know what you mean. Init img is just variables of text2image call.