Great video, thanks for sharing! Have you experimented with other LLMs than Zephyr for generating prompts? Just curious if maybe there are other 7B or 13B models that would give equal or better results.
Thanks for these videos - you are covering really interesting subjects that others are not and I am experimenting with the things you are demonstrating and having fun - keep it up👌
I've just downloaded the png image for this workflow and tried dragging it into ComfyUI, but the workflow isn't embedded in it - I've checked it by opening the PNG in notepad++ - the images with embedded workflow have plain text in the header that is interpreted by Comfy, but your images haven't got that text in, so Comfy cannot convert them to workflow (the same issue occurs for the img to img added noise workflow as well). Did you do anything with the PNGs generated by Comfy within another image software package before uploading them, as this might have stripped out the text content? If so, could you upload the images directly exported from Comfy, which will retain the neccessary data? Thanks
@@dcpuzzles2990 actually I did and I made sure that WordPress will not run over it. Of there's a better way to serve these files, please let me know... I'll be happy to use it.
I was considering an LLM but now I'm a bit puzzled, the prompt it made was quite poor. No red jacket... a sort of cape, the blue from the eyes bled everywhere, no anime, no lute guitar instead, a cathedral not a castle, a rock star not a bard... it took me five min to write a shorter weighted prompt which delivered all of these. Does it have other settings?
Actually it has. But I guess the results are more related to the SD model art style and not the llm model, and since it's a matter of artistic preference, ymmv for sure. I would still suggest playing with it, you can always delete it if you don't like what you get
none taken, i did not develop the LLM :) i just think it's a cool tool to have - I guess with better prompting and some hacking to the node you can make it to your standards. if you do that, Please let me play with it :-)
I wish I was clever enough! I guess it will get better very fast, I'd expected it to use weights like (lute:1.4) which will make it a lute rather than a guitar, or (blue eyes:0.4) which would stop the blue spreading too much.@@Just_gili
That’s amazing. Love the roller coasting chick
Great video, I will experiment with this !
thanks, let us know what you've came up with
Great video, thanks for sharing! Have you experimented with other LLMs than Zephyr for generating prompts? Just curious if maybe there are other 7B or 13B models that would give equal or better results.
Actually not yet. I'm quite happy with these results. I do wonder if I can refine the prompt though - the ones that embedded within the node.
Thanks for these videos - you are covering really interesting subjects that others are not and I am experimenting with the things you are demonstrating and having fun - keep it up👌
Thanks! I'm showing the stuff that i've playing with and doing a video about these is making me understand it better. so it's a win win i guess.
I've just downloaded the png image for this workflow and tried dragging it into ComfyUI, but the workflow isn't embedded in it - I've checked it by opening the PNG in notepad++ - the images with embedded workflow have plain text in the header that is interpreted by Comfy, but your images haven't got that text in, so Comfy cannot convert them to workflow (the same issue occurs for the img to img added noise workflow as well). Did you do anything with the PNGs generated by Comfy within another image software package before uploading them, as this might have stripped out the text content? If so, could you upload the images directly exported from Comfy, which will retain the neccessary data? Thanks
@@dcpuzzles2990 actually I did and I made sure that WordPress will not run over it.
Of there's a better way to serve these files, please let me know... I'll be happy to use it.
@@Just_gili could you upload the json scripts and put links for direct download? That would guarantee that the workflows can be used. Thanks
I was considering an LLM but now I'm a bit puzzled, the prompt it made was quite poor. No red jacket... a sort of cape, the blue from the eyes bled everywhere, no anime, no lute guitar instead, a cathedral not a castle, a rock star not a bard... it took me five min to write a shorter weighted prompt which delivered all of these. Does it have other settings?
Actually it has.
But I guess the results are more related to the SD model art style and not the llm model, and since it's a matter of artistic preference, ymmv for sure.
I would still suggest playing with it, you can always delete it if you don't like what you get
Maybe I was expecting too much, no offence, but the LLM prompt is badly constructed, I expect it will improve rapidly with training though.@@Just_gili
none taken, i did not develop the LLM :) i just think it's a cool tool to have -
I guess with better prompting and some hacking to the node you can make it to your standards. if you do that, Please let me play with it :-)
I wish I was clever enough! I guess it will get better very fast, I'd expected it to use weights like (lute:1.4) which will make it a lute rather than a guitar, or (blue eyes:0.4) which would stop the blue spreading too much.@@Just_gili
super thanks