yea these are very interesting for their use with inpainting , can imagine this as being more controlled underlying steps for things like omnigen manipulation . and more modularized as we can train specific synthetic transformations as seperate LoRAs then apply that any input image , its much more then a style transfer as is not limited by the underlying starting pixels of img2img
Thanks for your workflows. Two things: 1) Your workflow asks nodes called Primere. When I installed them and refreshed Comfy started dl'ing lots of checkpoints (GB's). I had to stop and delete Primere. 2) I tried to use "brand identity" 2-image panel to put a specific fabric pattern onto an existing image of a man wearing a jacket. I masked the jacket on the right image but neither Flux nor Flux Fill could put the pattern on that jacket. I take it's impossible right now to in-context/inpaint a masked area (jacket) in the right-hand image?
I don’t use and had never heard of “primere”, but after hunting it down it looks like it has some conflicting nodes. It should be using the ones as shown in the video. As manager can be touch and go, I turn node names on in my videos so it’s easy to see the correct packs. I’ve not tried masking with these loras, but that sounds like more of a job for IPAdapter or redux!
Hi - I am trying to use the StoryBoard Lora - and every time I am getting subtitles appear on one or more of the boards. I don't know if the lora was trained on video with subtitles or something, but I can't get rid of them. I tried adding in a dynamic thresholding solution, but that didn't really work. I am also using Flux Pulid to get a consistent character from reference shot - but that only makes the subtitling worse. - Any ideas?
Not sure why i cant get this to work, and every workflow ive found provided by these ppl, have 30 different redundant node packs to do the most simple things that can be done with base nodes, its insanely annoying.
This is one of the bigger issues with Comfy since it became popular - bloat. So many nodes get repeated because many do the same nodes over and over when it's not needed.. It really needs some form of consolidation or curation but I can't see many wanting to take it on.
@@NerdyRodent Hey. Hehe, ya slowly working out it. Seems to really need the prompt to be perfect in order to do the logo transfer etc. Love your videos
@@Elwaves2925 Ya it can be a mess at times. Then you have like some nodes that dont work with 'batches' of images, and some that do etc I guess it stems from the original comfyui, lacking some really key basic nodes for some reason, which resulted in so many different people making those basic nodes, and they all have massive packs of nodes that want tons of python libs etc
I am so glad you are covering this. In-Lora is very helpful
This works pretty dang good actually, thanks for exposing me to this.
I'll have to try these in the morning but I like the look of them.
Getting closer to consistency workflows for comics and animations.
Control net will get you there.
Nice tools and ideas. Thank you very much.
Love the videos always great content
Appreciate it!
Oh, Nerdy Rodent! 🐭🎵
He really makes my day! ☀
Showing us AI, 🤖
in a really British way! ☕🎶
😀
yea these are very interesting for their use with inpainting , can imagine this as being more controlled underlying steps for things like omnigen manipulation . and more modularized as we can train specific synthetic transformations as seperate LoRAs then apply that any input image , its much more then a style transfer as is not limited by the underlying starting pixels of img2img
💗 🐁
Is the outpaint workflow available somewhere? on your patreon?
It is indeed, along with a bunch of others! Supporters also get access before the video comes out too 😉
My lovely Rodent, thx 💫
Thanks for your workflows. Two things:
1) Your workflow asks nodes called Primere. When I installed them and refreshed Comfy started dl'ing lots of checkpoints (GB's). I had to stop and delete Primere.
2) I tried to use "brand identity" 2-image panel to put a specific fabric pattern onto an existing image of a man wearing a jacket. I masked the jacket on the right image but neither Flux nor Flux Fill could put the pattern on that jacket. I take it's impossible right now to in-context/inpaint a masked area (jacket) in the right-hand image?
I don’t use and had never heard of “primere”, but after hunting it down it looks like it has some conflicting nodes. It should be using the ones as shown in the video. As manager can be touch and go, I turn node names on in my videos so it’s easy to see the correct packs. I’ve not tried masking with these loras, but that sounds like more of a job for IPAdapter or redux!
very cool. reminds me of omnigen
Can this lora be used with Flux schnell as well?
Hi - I am trying to use the StoryBoard Lora - and every time I am getting subtitles appear on one or more of the boards. I don't know if the lora was trained on video with subtitles or something, but I can't get rid of them. I tried adding in a dynamic thresholding solution, but that didn't really work. I am also using Flux Pulid to get a consistent character from reference shot - but that only makes the subtitling worse. - Any ideas?
Not had that issue as yet myself!
@ after hours of experimenting, turned out all I had to do was change the seed lololol
can we use this in forge?
Does forge support flux & flux loras?
@@NerdyRodent yes it does
its possible to add my own lora ? like lora character
Yup!
5:21 unfortunatly the trick with other loras dosent work well, probably only with this design lora one do his job better no idea
we need more gguf flux :(
Couldn't find the proof of concept outpainting wf on patreon
It’s the second JSON file (with IMGOP in the name), as detailed under the “workflows” heading with more info
could you please share your workflow file here? the one with outpainting
Not sure why i cant get this to work, and every workflow ive found provided by these ppl, have 30 different redundant node packs to do the most simple things that can be done with base nodes, its insanely annoying.
This is one of the bigger issues with Comfy since it became popular - bloat. So many nodes get repeated because many do the same nodes over and over when it's not needed.. It really needs some form of consolidation or curation but I can't see many wanting to take it on.
Luckily, none of that applies here as you simply need the base Lora loader! 😃
@@NerdyRodent Hey. Hehe, ya slowly working out it. Seems to really need the prompt to be perfect in order to do the logo transfer etc.
Love your videos
@@Elwaves2925 Ya it can be a mess at times. Then you have like some nodes that dont work with 'batches' of images, and some that do etc
I guess it stems from the original comfyui, lacking some really key basic nodes for some reason, which resulted in so many different people making those basic nodes, and they all have massive packs of nodes that want tons of python libs etc
👋 hi
Howdy!
This would be nice as img2img node without the need to work around
this is Flux???? no thanks 3:15