lollms-webui icon indicating copy to clipboard operation
lollms-webui copied to clipboard

REQUEST Comfyui support via json workflow sending.

Open GamingDaveUk opened this issue 6 months ago • 0 comments

Currently you have automatic1111 loaded up and run inside of the personality creator which is impressive, but i propose something with some more versatility to it, that I am not sure if its do-able:

ComfyUI: https://github.com/comfyanonymous/ComfyUI is an intimidating to look at AIArt creation system but is actually pretty easy to use and allows for some real complexities. It also supports recieving "workflows" as jsons via its api, this is how silly tavern uses it, theres a document on how a chap used this to do batch processing: https://blog.xentoo.info/2023/07/22/comfyui-batch-run-from-command-line-with-api/#:~:text=When%20you%20click%20%E2%80%9Cqueue%20prompt%E2%80%9D%20in%20ComfyUI%2C%20it,now%20see%20a%20new%20button%20Save%20%28API%20format%29.

Now with a workflow json the prompt for positive and negative is included as is the model to load, steps, cfg etc could these be made wild cards ie: positivePrompt negativePrompt cfg steps model height width

and then have the ai replace those wild cards in the prompt?

So say we prompted a comfyui art persona: "Give me an image of a setting sun, over a town, make it portrait and use model "SDXL_realistic" " The persona would; create an art prompt based on "setting sun, over a town" and put that in the positivePrompt wildcard. realise there is no negative prompt so blank that (with sdxl you really dont need negatives 90% of the time) realise there is no cfg or steps so use the default we have given it. notice we want portrait so use the height and width we have set for portrait.

is that possible? and if so can we have it? ideally the workflow templates should be stored in a folder so we can add our own. I have multiple work flows I use and it would be easy to modify them for wildcards as needed.

There is a site dedicated to sharing workflows if you want to see some more advanced ones https://openart.ai/workflows/dev but I would recommend simply including the default workflow for comfyui for maximum compatability. who knows one day maybe ai's will be smart enough to write a workflow in full for our needs as we need them lol

For example I have a workflow that takes in an image and does essentially what automatic111 sd upscale script does, being able to give a bot that image and say "send this to the upscale workflow" could be really handy

GamingDaveUk avatar Dec 19 '23 20:12 GamingDaveUk