to make the workflow even cleaner, use the aux preprocessor from the ComfyUI's ControlNet Auxiliary Preprocessors node which has all the preprocessors you need!
the only problem i see is that you lose some of the fine tuning for each preprocessor. like with the standalone canny, you can adjust the low and high thresholds, but with the aux, you dont have that control.
Thanks for the video! Well presented and straight to the point, very practical. It would be interesting to see some testing daisy chaining controlnet, if that even works on Flux, combining lets say depth and canny could be useful in certain scenarios.
detailed tutorial thank you. I've come to the same conclusion about flux ControlNets. XLabs are not usable with a proper workflow, and the same applies to Mistoline ControlNet, which I've found to be superior to Canny. However, we must use a specific KSampler so same problem as xlabs..
Yeah but I don't mind. I really like the FluxSamplerParams sampler anyway. If you haven't heard of that one, check the testing video. That sampler is awesome. th-cam.com/video/EZ9btjSkNR4/w-d-xo.htmlsi=OtMiZmb5wVAy5va4
I tried using mistoline and xlabs depth controlnet with CR multi-controlnet stack node with CR apply multi-controlnet. It worked, but I did not compare this setup with stand alone controlnet setup.
I will have another Control Net video coming soon that will blow your mind. Join the streams sometime and I'll take your requests for anything you want to try. And get in the discord if you need help, have suggestions and/or want to chat and share you images for the weekly contest. discord.gg/KgxgXBjMX6
i didnt know any of the samplers need 1:1? i havent messed with the ksampler in a while so i dont remember. i know the flux sampler params can do any ratio ive tried, which is a ton
@@flowstateeng In your video you're using 1024x1024 generation. Try loading a non-1:1-ratio-image and see if it works. For me your workflow only worked with square images.
@kallamamran oh yeah with the control nets it is a little weird but that is the apply control net node. I'm not really sure what's going on there. It's not strictly square images but it's very picky about the values it likes for other ratios. It's not the sampler though for sure. I do all my thumbs in 1080 with that sampler
yeah its probably just an incompatibility between that CN apply and the flux params sampler. im building some custom nodes now, so im sure ill be able to see whats going on there when i dig in to that part of the code
yeah its not 100% perfect but still the best option ive found. i actually am recording a video right now bc i dove into making my own nodes, so i may take a look at creating a new version of flux sampler params that fixes that. i already posted the python file with my first node that im making a video about now. its on the patreon and on git if you wanna grab it. if i do make the new sampler, i will put it in that file and update the patreon post. trying to keep all my nodes in that one file to make things easy
Hey can you check my new vid and lmk what you think of the audio? It's a long one but if you can just listen for a minute or two and lmk that would be really helpful. I tweaked some mic settings and I also boosted it in post. I was worried that it would make the S's and P's too harsh, which is why it was quieter before but I think it actually turned out pretty well. Would be good to get another set of ears on it though to be sure. Thanks for the heads up on the audio! th-cam.com/video/Gsx8WtyQj2Q/w-d-xo.htmlsi=lBIzpxaq5YBm41M3
@@flowstateeng The new video mic audio sounds even slightly more mushy than this video. It sounds like low frequencies are way too loud, like it fills my head when listening with headphones. I'm not audio guy, but I watch/listen a ton of YT learning material. I've got something like twenty tabs of similar content from different youtubers open right now, and I compared the audio. Most if not all, have "thinner" voice in their video. Not like other's have it so that it sounds like a low quality phone call, but I hope you get the idea. I don't like the idea of mentioning other channels, but I guess it might help you here, I can remove the message if you want. For example, SECourses has often bit low quality audio, but his videos are easier on ears, as there is less loud low frequencies in voice. Same with CyberJungle, pretty similar in quality, but way less powerful low frequencies in voice so quite easy to listen. Channels like PixelEasel who use AI speech are very easy on ears, as the voice sounds relatively thin and neutral (talking about audio quality). Or maybe simply check big youtubers on AI, like Theoretically Media, his videos have really natural voice quality. I don't have high quality headphones, but like I said, I listen hundreds of hours of YT videos each year (hours each day usually), and your mic sound feel like its settings are really bad, compared to majority. Maybe rip/download some of these videos, and see how your voice differs in audio/video editing software? It is not impossible to listen to your videos, but it sounds like my head was inside a small box that somehow amplifies lower frequencies of your voice, that soft and muffled part is really overwhelming, and 95% of YT channels don't have that. I remember watching early Brackeys videos (Unity youtuber) those had really bad muffled audio. Even if I swap between your latest video, and Theoretically media's latest video, on his video I can move the volume knob to max and his voice still doesn't start to feel bad on my ears.
And to be 100% clear, I don't say these comments because I want you to fail, quite the opposite, I enjoy high quality content, and audio quality is most often the reason I stop following people even if the content was otherwise nice. Like some Indian tech youtubers, I don't care about the accent, but that low quality mic with static noise combined with background noises (for example) is too much.
Yeah you're good, I definitely appreciate the feedback! I don't take it negatively at all. The goal is for the content to be as good as possible. If it's bad people won't watch! So any feedback is great because it'll help me make things better for you guys! I'm definitely a novice on the A/V side of things, so any tips at all help a lot. Thanks for the mentions of other channels. I will compare the audio and see what I can do to tune it more like theirs.
@@flowstateeng I'm no audio guy, just hear from my buddy in work about these things. But try to find videos on how to setup mic and how to fix common audio quality issues. Of course, it is best to have high quality capture, like with photos, it is hard to fix bad quality in post, if it is even possible for acceptable quality. Also, for example, being too close to a mic may also cause similar issues (causes distortion). And some types of microphones simply give different kind of tone for your recording, see which ones are recommended I guess, if everything else fails.
Thanks great video. Love the explanation and clean workflow.
I really hope flux gets a lot more controlnets. It really produces some great quality.
thanks! yeah they will soon.
to make the workflow even cleaner, use the aux preprocessor from the ComfyUI's ControlNet Auxiliary Preprocessors node which has all the preprocessors you need!
hey yeah i stumbled across that this morning. I'm checking it out now to make sure it works with everything. thanks for the heads up!
the only problem i see is that you lose some of the fine tuning for each preprocessor. like with the standalone canny, you can adjust the low and high thresholds, but with the aux, you dont have that control.
@@flowstateeng from my experience, you can adjust the resolution in aux, which gives similar fine tuning result.
cool ill check it out, thanks!
Thanks for the video! Well presented and straight to the point, very practical. It would be interesting to see some testing daisy chaining controlnet, if that even works on Flux, combining lets say depth and canny could be useful in certain scenarios.
yep, thats exactly what i was thinking. im working on some more controlnet and ipadapter vids now doing exactly that!
detailed tutorial thank you. I've come to the same conclusion about flux ControlNets. XLabs are not usable with a proper workflow, and the same applies to Mistoline ControlNet, which I've found to be superior to Canny. However, we must use a specific KSampler so same problem as xlabs..
Yeah but I don't mind. I really like the FluxSamplerParams sampler anyway. If you haven't heard of that one, check the testing video. That sampler is awesome. th-cam.com/video/EZ9btjSkNR4/w-d-xo.htmlsi=OtMiZmb5wVAy5va4
@@flowstateeng yeah i'm using it with a custom version of your own workflow =D it's really good indeed
I tried using mistoline and xlabs depth controlnet with CR multi-controlnet stack node with CR apply multi-controlnet. It worked, but I did not compare this setup with stand alone controlnet setup.
yeah im gonna do some vids on multi control net stacks with this workflow tomorrow. looks really useful
@@john2stai eeeh... Mistoline is SDXL huh?! 🤔
nice! would like to see these at lower guidance say 3.5 which might make it less "hyperreal"?
I will have another Control Net video coming soon that will blow your mind. Join the streams sometime and I'll take your requests for anything you want to try. And get in the discord if you need help, have suggestions and/or want to chat and share you images for the weekly contest. discord.gg/KgxgXBjMX6
Xlabs Sampler doesn't need 1:1 input latents, so it's more flexible regarding input and output ratios
i didnt know any of the samplers need 1:1? i havent messed with the ksampler in a while so i dont remember. i know the flux sampler params can do any ratio ive tried, which is a ton
@@flowstateeng In your video you're using 1024x1024 generation. Try loading a non-1:1-ratio-image and see if it works. For me your workflow only worked with square images.
@kallamamran oh yeah with the control nets it is a little weird but that is the apply control net node. I'm not really sure what's going on there. It's not strictly square images but it's very picky about the values it likes for other ratios. It's not the sampler though for sure. I do all my thumbs in 1080 with that sampler
@@flowstateeng Hmmm... I have had no problem with the CN and size before. Just keeping the image dimensions divisible by 8 and it's usually fine
yeah its probably just an incompatibility between that CN apply and the flux params sampler. im building some custom nodes now, so im sure ill be able to see whats going on there when i dig in to that part of the code
Flux Sampler Parameters seed is string formatted though. That's not good. Makes it hard using "Seed everywhere"
yeah its not 100% perfect but still the best option ive found. i actually am recording a video right now bc i dove into making my own nodes, so i may take a look at creating a new version of flux sampler params that fixes that. i already posted the python file with my first node that im making a video about now. its on the patreon and on git if you wanna grab it. if i do make the new sampler, i will put it in that file and update the patreon post. trying to keep all my nodes in that one file to make things easy
im also gonna be making my own versions of the use everywhere nodes bc there are some other issues ive had with that too
Otherwise OK video, but your audio / voice is way to soft and mushy.
Hey can you check my new vid and lmk what you think of the audio? It's a long one but if you can just listen for a minute or two and lmk that would be really helpful. I tweaked some mic settings and I also boosted it in post. I was worried that it would make the S's and P's too harsh, which is why it was quieter before but I think it actually turned out pretty well. Would be good to get another set of ears on it though to be sure. Thanks for the heads up on the audio!
th-cam.com/video/Gsx8WtyQj2Q/w-d-xo.htmlsi=lBIzpxaq5YBm41M3
@@flowstateeng The new video mic audio sounds even slightly more mushy than this video. It sounds like low frequencies are way too loud, like it fills my head when listening with headphones. I'm not audio guy, but I watch/listen a ton of YT learning material. I've got something like twenty tabs of similar content from different youtubers open right now, and I compared the audio. Most if not all, have "thinner" voice in their video. Not like other's have it so that it sounds like a low quality phone call, but I hope you get the idea. I don't like the idea of mentioning other channels, but I guess it might help you here, I can remove the message if you want. For example, SECourses has often bit low quality audio, but his videos are easier on ears, as there is less loud low frequencies in voice. Same with CyberJungle, pretty similar in quality, but way less powerful low frequencies in voice so quite easy to listen. Channels like PixelEasel who use AI speech are very easy on ears, as the voice sounds relatively thin and neutral (talking about audio quality). Or maybe simply check big youtubers on AI, like Theoretically Media, his videos have really natural voice quality. I don't have high quality headphones, but like I said, I listen hundreds of hours of YT videos each year (hours each day usually), and your mic sound feel like its settings are really bad, compared to majority. Maybe rip/download some of these videos, and see how your voice differs in audio/video editing software? It is not impossible to listen to your videos, but it sounds like my head was inside a small box that somehow amplifies lower frequencies of your voice, that soft and muffled part is really overwhelming, and 95% of YT channels don't have that. I remember watching early Brackeys videos (Unity youtuber) those had really bad muffled audio. Even if I swap between your latest video, and Theoretically media's latest video, on his video I can move the volume knob to max and his voice still doesn't start to feel bad on my ears.
And to be 100% clear, I don't say these comments because I want you to fail, quite the opposite, I enjoy high quality content, and audio quality is most often the reason I stop following people even if the content was otherwise nice. Like some Indian tech youtubers, I don't care about the accent, but that low quality mic with static noise combined with background noises (for example) is too much.
Yeah you're good, I definitely appreciate the feedback! I don't take it negatively at all. The goal is for the content to be as good as possible. If it's bad people won't watch! So any feedback is great because it'll help me make things better for you guys! I'm definitely a novice on the A/V side of things, so any tips at all help a lot. Thanks for the mentions of other channels. I will compare the audio and see what I can do to tune it more like theirs.
@@flowstateeng I'm no audio guy, just hear from my buddy in work about these things. But try to find videos on how to setup mic and how to fix common audio quality issues. Of course, it is best to have high quality capture, like with photos, it is hard to fix bad quality in post, if it is even possible for acceptable quality. Also, for example, being too close to a mic may also cause similar issues (causes distortion). And some types of microphones simply give different kind of tone for your recording, see which ones are recommended I guess, if everything else fails.