Text-to-image local model
Hello, i'm brand new to any of this.
I have zero programming skills all is done by or with ai.
Long read ahead
soft/hardware:
7900xt
9850x3d
32g ddr5 ram
Open Webui, comfyui, LM Studio Bionic.
Model: Qwen3.5-9b for chatting in open-webui.
I've currently got a Qwen-image-edit workflow working locally and editing images per my prompt then allows me to view or save the image.
The past two days i've spent all day with the free versions of chat-gpt and grok. Just yesterday i figured i'd grab the work laptop with a subscription for co-pilot as well.
What i'd like to achieve:
A chat-gpt like environment (open webui serves perfectly fine i think) that allows me to chat, ask questions, do internet searches with, upload images to and have them edited per my prompt as closely to the real picture as possible, and eventually maybe video/music editing (i believe this may ask too much of my system?)
So pretty much, chat-gpt but fully local on my pc.
Base history:
I wish not to pay for any of this so set out to 'make' one with the free tools.
With chat-gpt, grok and co-pilot i managed to make my pc AI ready with ROCm and start my comfyui environment with a .bat file including 'venv' and the comfy manager (apologies if what i write makes little sense). 'everything' seems to work.
I started on the text-to-image workflow first as it seemed the simplest. It's as of yet the only workflow i've been trying to get to work.
The text-to-image workflow is almost complete, i'm just missing a vital component. The Load image from api node.
I can currently upload an image in the comfyui environment Load Image node and have it edited per my prompt, then view and/or save it (view image). But i cannot get the Open Webui to work like this.
Many hours of troubleshooting later i figured out that (noob failulre) the Base64 node is 2 years old and outdated/unusable.
Second, the Load Image node in ComfyUI cannot load images from api. (what i mean by that is i want it to take the image i upload into Open Webui chat and carry out the prompt i send it with, then receive it back in the same chat).
On top of that the Load Image node cannot be left empty. It Requires a media file to be physically uploaded to it's node.
Now i stumbled across the Acly tooling nodes here:
Acly/comfyui-tooling-nodes
and slapped the documentation text into co-pilot (co-pilot cant search in the link, work forbade that).
Copilot says:
The Acly documentation explicitly says:
>Load Image from Cache requires an image ID that was previously uploaded via PUT /api/etn/image/{id}.
That means Acly itself does not magically receive images from Open WebUI. Something must still:
1. Receive the uploaded image.
2. Upload it to the Acly cache endpoint.
3. Generate or know the cache ID.
4. Pass that cache ID into the node.
The documentation does not state that Open WebUI performs those steps automatically. [https://do...hatgpt.com]
Therefore I cannot confirm that replacing your Load Image node with Acly Load Image from Cache will work in Open WebUI without an additional integration layer.
So is there really no current option at all to do what i wish? Is there no such node to convey the uploaded image from open webui into the comfyui workflow?
Any tips are immensely appreciated. If any more files/info is required i'll gladly upload them
Edit: SOLVED
3 days of troubleshooting came down to a damned capital "i" in the Open webui > images > node id's.....
In the top row i filled in 'Images' but the .py script that looks at that node tries to read 'images'
Thats why it failed to take my uploaded image through the comfyui workflow. Not sure if i should or shouldn't be glad i never became a programmer lol
9 comments
Mentions
Text-to-image local model Hello, i'm brand new to any of this. I have zero programming skills all is done by or with ai. Long read ahead soft/hardware: 7900xt 9850x3d 32g ddr5 ram Open Webui, comfyui, LM Studio Bionic. Model: Qwen3.5-9b for chatting in open-webui. I've currently got a Qwen-image-edit workflow working locally and editing images per my prompt then allows me to view or save the image. The past two days i've spent all day with the free versions of chat-gpt and grok. Just yesterday i figured i'd grab the work laptop with a subscription for co-pilot as well. What i'd like to achieve: A chat-gpt like environment (open webui serves perfectly fine i think) that allows me to chat, ask questions, do internet searches with, upload images to and have them edited per my prompt as closely to the real picture as possible, and eventually maybe video/music editing (i believe this may ask too much of my system?) So pretty much, chat-gpt but fully local on my pc. Base history: I wish not to pay for any of this so set out to 'make' one with the free tools. With chat-gpt, grok and co-pilot i managed to make my pc AI ready with ROCm and start my comfyui environment with a .bat file including 'venv' and the comfy manager (apologies if what i write makes little sense). 'everything' seems to work. I started on the text-to-image workflow first as it seemed the simplest. It's as of yet the only workflow i've been trying to get to work. The text-to-image workflow is almost complete, i'm just missing a vital component. The Load image from api node. I can currently upload an image in the comfyui environment Load Image node and have it edited per my prompt, then view and/or save it (view image). But i cannot get the Open Webui to work like this. Many hours of troubleshooting later i figured out that (noob failulre) the Base64 node is 2 years old and outdated/unusable. Second, the Load Image node in ComfyUI cannot load images from api. (what i mean by that is i want it to take the image i upload into Open Webui chat and carry out the prompt i send it with, then receive it back in the same chat). On top of that the Load Image node cannot be left empty. It Requires a media file to be physically uploaded to it's node. Now i stumbled across the Acly tooling nodes here: Acly/comfyui-tooling-nodes and slapped the documentation text into co-pilot (co-pilot cant search in the link, work forbade that). Copilot says: The Acly documentation explicitly says: >Load Image from Cache requires an image ID that was previously uploaded via PUT /api/etn/image/{id}. That means Acly itself does not magically receive images from Open WebUI. Something must still: 1. Receive the uploaded image. 2. Upload it to the Acly cache endpoint. 3. Generate or know the cache ID. 4. Pass that cache ID into the node. The documentation does not state that Open WebUI performs those steps automatically. [https://do...hatgpt.com] Therefore I cannot confirm that replacing your Load Image node with Acly Load Image from Cache will work in Open WebUI without an additional integration layer. So is there really no current option at all to do what i wish? Is there no such node to convey the uploaded image from open webui into the comfyui workflow? Any tips are immensely appreciated. If any more files/info is required i'll gladly upload them Edit: SOLVED 3 days of troubleshooting came down to a damned capital "i" in the Open webui > images > node id's..... In the top row i filled in 'Images' but the .py script that looks at that node tries to read 'images' Thats why it failed to take my uploaded image through the comfyui workflow. Not sure if i should or shouldn't be glad i never became a programmer lol