Beauties and Beasts (Comfyui/Gligen Workflow)
Beauties and Beasts
FeelYou has given me some very nice results messing around with some of the example prompts. Something I appreciate is that you can get good results with simple prompts without resorting to what I call 'word salad' prompts. Bradcatt's models are also always ones I tend to come back to and I really like the colors of FeelYou and SeekYou in particular, so thank you, Bradcatt, for your time and effort in making these models and sharing example images for us to learn from.
In that spirit, I wanted to give something back. I made the following in Comfyui using FeelYou Alpha 2 and the add_detail lora in order to test out and learn how gligen textboxes worked.
Resources used:
FeelYou Alpha 2
Workflow Explanation:
Gligen boxes are bounding boxes defined in pixels that, to my understanding, allow you to take a concept from your positive prompt and define where it is in the image and further describe/enhance the text for that concept without having those details bleed into other areas of the image.
The Comfyui implementation prefers dimensions divisible by 4 pixels. Each Gligen Text Box in Comfyui has fields for the prompt and the bounding box dimensions and starting coordinate (the upper left corner of the bounding box with 0,0 being the upper left of the whole image.)
This is a screenshot of my latest node layout for this. Drag and drop one of the later images in this post (probably the last one in this post for the most update/clean version) to get a copy of this node layout into Comfyui. The Comfui Manager should prompt you to install any missing custom nodes and it will download them for you.
My Favorite Results So Far:
Most of my images had two gligen bounding boxes, one for a woman character and another for a monster character, but you can have more if needed. I'm still learning prompt weighting in Comfyui but here are some of my favorite results so far from my experimenting with gligen.
Images 1-3 have the monster gligen box equal to the whole image. The monster ended up fairly large like I intended but it doesn't always fill the whole space and is often not fully in the frame. I later figured out that moving the edges of the gligen box off of the outer edges of the image would force the concept to stay in frame. The woman gligen box was a rectangle reaching up from the center of the bottom edge and was often a medium shot, again most likely because the gligen box is touching the bottom edge.
Images 4-5 have the woman gligen box in the lower right hand corner and this tended to result in more closeup shots of that character I think because it touched two outer edges.
In Images 6-9 have gligen boxes of equal size, dividing the image into equal halves vertically. Mostly closeups, again probably because the gligen boxes are touching three outer edges.
Images 10-14 returned to the gligen setup I started with, though images 10-12 with the wolves also added a third gligen box for a moon in the upper left corner. That gligen box didn't touch any outer edges and you can see that the moon is fairly consistent in size and location. Because I referred to 'moon' in the main prompt, there was occasionally a second large moon somewhere in the image.
If you made it this far, thanks for reading and happy rendering.