Thanks, glad you liked it! I'm diving way deeper into IP Adapters, Control Nets, and Stable Video Diffusion in my next few vids, so there's much more to come on that
hey thanks! its not necessary to add but it does help produce good results in some cases. you can set the values all to 1, if you want to disable it, or remove it entirely from the workflow. also make sure to get in the discord! we got a nice community over there to help with bugs, questions, share techniques, etc. its coming along nicely! discord.gg/KgxgXBjMX6
can you hop in the discord and ask in the bug chat! we got a nice community over there to help with bugs, questions, share techniques, etc. makes it easier to share screenshots of your CMD output, etc. discord.gg/KgxgXBjMX6
How about a video on posing existing characters? Not sure if flux ipadapter has an openpose to be able to do something other than style transfer. I've been unsuccessfully searching for a method to retain characteristics of existing character and change their pose.
I'm still going to do this with the IP Adapter and Control Net but just wanted to mention, one effective way to do this is to simply load your image of your existing character, pass that into a VAE Encode node, and send that into your sampler as your latent, instead of using an empty latent image. I've been playing around with this a lot lately and its every effective at transferring the image content (including the pose of any characters in the image) and style to the generated image. You just vary your denoise value to control the effect. Lower denoise == more of your input image, higher denoise == more of your generated image. Around 0.85 - 0.95 is a good range for the denoise. I show some examples in the video below. I'll add a comment here when I post the videos covering the IP Adapter & Control Net methods for doing it though. th-cam.com/video/5wAHg8OvgbM/w-d-xo.htmlsi=-DVPethuqaZVJ7f3
It helps to have a prompt that's somewhat similar to the shape of your input image. For example, if your input is a person, it can be any human-like object. I did one at the end of my new video where the input was a close up shot of a cyberpunk woman on a street, and the prompt was "a close up shot of a minotaur in a china shop." The lower the denoise, the more woman-like the minotaur becomes, and the more the china shop begins to resemble a street with a building beside him and neon signs behind him. At 0.89 on the denoise I got a good mixture of the input and generated image. It was some kind of angry looking beast, in a china shop, in the same pose as the woman, with a wall to his right and one behind him like the input, and two signs in the background roughly in the same position as the two in the input but with different text, and very similar color palette as the input. Its a long vid, so here's the timestamp to that part: th-cam.com/video/Gsx8WtyQj2Q/w-d-xo.htmlsi=KCzBuS6XcPBx8hZ5&t=1543
did you open in a new tab? idk whats up with patreon but you gotta click the image, then right click and open in a new tab. that one will let you import the workflow. for some reason on patreon if you just drag the image in without opening in a new tab first, comfy doesnt recognize it
I had to click on the image so it opens in its own window then right click save image as to save to desktop, to get one to then be able to drag to comfy with the workflow embedded. Will try without saving next time.
yeah its because patreon makes image copies for the front end. the original image is on their server but what they show you in the browser is a .webp copy bc its smaller and more efficient for the browser to load. so if you click on the image, then right click and open that image in a new tab, you will get the original image in your new tab. or you can download it like @sven1858 said. i try to be explicit about those step in my videos but sometimes i breeze through it. soon all of the workflows will just be included in my custom nodes but we gotta deal with patreon for the next couple days
![Create a stunning eco-resin tray with silkscreen printing! [24]](http://i.ytimg.com/vi/MjRmcUTtUrA/mqdefault.jpg)
Nice job
Thanks, glad you liked it! I'm diving way deeper into IP Adapters, Control Nets, and Stable Video Diffusion in my next few vids, so there's much more to come on that
thanks for the tutorial... what is the point of adding the clip attention multiply node? is that necessary?
hey thanks! its not necessary to add but it does help produce good results in some cases. you can set the values all to 1, if you want to disable it, or remove it entirely from the workflow.
also make sure to get in the discord! we got a nice community over there to help with bugs, questions, share techniques, etc. its coming along nicely! discord.gg/KgxgXBjMX6
where does the ipadapter file go into which comfyui folder - models/ipadapter ? it is not showing up in the node
you'll want to put that one in the models/xlabs/ipadapters. the folder is created when you install their nodes
i have this error
FluxSamplerParams+
'ForgeParams4bit' object has no attribute 'quant_storage'
I make sure I followed the instructions more than once
can you hop in the discord and ask in the bug chat! we got a nice community over there to help with bugs, questions, share techniques, etc. makes it easier to share screenshots of your CMD output, etc.
discord.gg/KgxgXBjMX6
@@flowstateeng great community 👌
How about a video on posing existing characters? Not sure if flux ipadapter has an openpose to be able to do something other than style transfer. I've been unsuccessfully searching for a method to retain characteristics of existing character and change their pose.
yep, im actually working on that now. probably in the next 2-3 days. got a couple other things ahead of it
one way or another, im pretty sure i can get it worked out
I'm still going to do this with the IP Adapter and Control Net but just wanted to mention, one effective way to do this is to simply load your image of your existing character, pass that into a VAE Encode node, and send that into your sampler as your latent, instead of using an empty latent image. I've been playing around with this a lot lately and its every effective at transferring the image content (including the pose of any characters in the image) and style to the generated image. You just vary your denoise value to control the effect. Lower denoise == more of your input image, higher denoise == more of your generated image. Around 0.85 - 0.95 is a good range for the denoise. I show some examples in the video below. I'll add a comment here when I post the videos covering the IP Adapter & Control Net methods for doing it though.
th-cam.com/video/5wAHg8OvgbM/w-d-xo.htmlsi=-DVPethuqaZVJ7f3
@@flowstateeng appreciate this very much. I'll take a stab at it tomorrow. So no prompt to pose?
It helps to have a prompt that's somewhat similar to the shape of your input image. For example, if your input is a person, it can be any human-like object. I did one at the end of my new video where the input was a close up shot of a cyberpunk woman on a street, and the prompt was "a close up shot of a minotaur in a china shop." The lower the denoise, the more woman-like the minotaur becomes, and the more the china shop begins to resemble a street with a building beside him and neon signs behind him. At 0.89 on the denoise I got a good mixture of the input and generated image. It was some kind of angry looking beast, in a china shop, in the same pose as the woman, with a wall to his right and one behind him like the input, and two signs in the background roughly in the same position as the two in the input but with different text, and very similar color palette as the input.
Its a long vid, so here's the timestamp to that part: th-cam.com/video/Gsx8WtyQj2Q/w-d-xo.htmlsi=KCzBuS6XcPBx8hZ5&t=1543
I tried dragging the image into comfy but no workflow loads
did you open in a new tab? idk whats up with patreon but you gotta click the image, then right click and open in a new tab. that one will let you import the workflow. for some reason on patreon if you just drag the image in without opening in a new tab first, comfy doesnt recognize it
i just edited the post though, so the workflow file is there too now if the image wont work
I had to click on the image so it opens in its own window then right click save image as to save to desktop, to get one to then be able to drag to comfy with the workflow embedded. Will try without saving next time.
yeah its because patreon makes image copies for the front end. the original image is on their server but what they show you in the browser is a .webp copy bc its smaller and more efficient for the browser to load. so if you click on the image, then right click and open that image in a new tab, you will get the original image in your new tab. or you can download it like @sven1858 said. i try to be explicit about those step in my videos but sometimes i breeze through it. soon all of the workflows will just be included in my custom nodes but we gotta deal with patreon for the next couple days