ComfyUI Product Photography Inpainting workflow #comfyui #controlnet #ipadapter #workflow

hi today we will see how to take a pack shot of a real product and build around it an environment that reacts to it whether it is with Shadow lighting reflection or anything else this workflow is based on many tools that we have seen in the previous videos as usual we will start at the beginning I work with the prompt Styler we may want to combine different styles in the future and to make it convenient to change the prompt according to the product and the desired background I have created here two input texts for the product and the background that with the help of the find and replace node will be combined with the required style you can see here the finished prompt that was sent to clip and code I work with jugut lightning but you should check which checkpoints suits you best according to the required result all of this goes to the case sampler with settings that match the model we have chosen and we will also go over the control net let's bypass this group here I load the image of the product and because there are images with different sizes I also connect the image resize and adjust it to the required resolution you should pay attention not to change the proportions for this there is also a preview node here and of course pay attention to the quality of the source image because it can greatly affect the final result this way of working will also ensure us that the composition and proportions will be compatible between the result and the source after the resize we reach the segment anything I will leave a link to the Project's GitHub page in the description this tool allows us to create a mask and disconnect the object from the original environment according to a keyword so I simply turned the prompt into input and added to it the word that describes our product and here you see the mask that the anything segment produces we want to create an environment around the shoe and not change the product so I invert the mask with invert mask I connect the mask and the image to AV encoder for in painting and this is the result we got really not bad but as you can see the shoe is not exactly the same as the original image to help the model to accurately position and shape the original shoe I use control net because the composition is important it should be noted that we connect the image to the pre-processor after the resize I work with depth anything a model adapted to dsxl and here you can see the depth map we will click on Q we are on a fixed seed and the other parameters are the same so we can see the effect that the control net has on the result and this is the picture we received with the control net as you can see it is already much closer to the original in terms of composition and proportions but if we enlarge the pictures we can clearly see that the quality of the shoe is completely different note that it is no longer possible to understand what is written here in the text and the quality is not the same for the relatively high quality product image because we built the workflow correctly we will be able to fix it quite easily all we need is to use the shoe from the original photo I also added here the Llama remove object node that will clean the Bad shoe from the background we created and for the Llama to work better I also enlarg the mask a bit and this is the picture after removing the background we will connect this image to the original shoe using the image composite mask node to which we will also connect the mask from any depth and so that the connection is as clean as possible I passed the mask through the mask feather and dilate mask nodes which should be changed and adjusted according to the nature of the product and the quality of the original image and this is the result after putting everything together please note that it is not 100% perfect there are all kinds of small disturbances here which can be fixed in a few seconds in Photoshop or but you will see that we managed to maintain the quality of the original product and the reflection Blends in nicely with the shoe and the background and of course this is only the starting point you can try to make combinations with backgrounds and different styles and for this I also added here the IP adapter Gara V2 of course which will allow you to use the image as a basis for the style you want to contain on the background so I hope you learned and we will meet in the next lessons you are of course welcome to subscribe to the channel ask questions and like if you liked it and most importantly have fun bye

Channel: PixelEasel

Views: 3,374

Keywords:

Id: k76f8aVgS4c

Length: 4min 32sec (272 seconds)

Published: Fri Apr 12 2024