MultiDiffusion: Fusing Diffusion Paths for Controlled Image Generation

Supplementary Material

Results and comparisons:
- Region-based generation
- Panorama

We recommend viewing all content in full screen. Click on the images for viewing them in full size.

Region-based generation

We provide additional results of our method on generation from rough scene layouts. For each spatial layout, we show different samples from our method.

	"a photo of a meadow" "an elephant" "a cow"


	"a photo of a desert" "a camel" "a cactus"

	"a photo of a beach" "a horse" "a chair"

	"blurred image" "a small bird" "a tree trunk"


	"blurred image" "a mouse" "a pile of books"

"a photo" "a balloon" "the Eiffel tower" "a pink cloud"

"a photo of the dolomites" "a robot" "a horse" a beautiful sunset

"falling stars in space" "a rocket" "blue smoke" "pink smoke"

"a photo of a beach" "a palm tree" "an air balloon"

"a photo of the dolomites" "a robot" "a lake" "a beautiful sunset"

Panorama Comparison

Sample results of our method on a variety of panorama images, at resolution 512x4608. Compared to Blended Latent Diffusion (combined with Stable Diffusion), Stable Inpainting, and ours

	A photo of a city skyline at night
	Blended+Stable

	Stable Inpainting

	Ours

	A photo of lush forest with a babbling brook
	Blended+Stable

	Stable Inpainting

	Ours

	A photo of a forest with a misty fog
	Blended+Stable

	Stable Inpainting

	Ours

	A photo of mountain range at twilight
	Blended+Stable

	Stable Inpainting

	Ours

	Cartoon panorama of spring summer beautiful nature
	Blended+Stable

	Stable Inpainting

	Ours

	A photo of a snowy mountain peak with skiers
	Blended+Stable

	Stable Inpainting

	Ours

	Natural landscape in anime style illustration
	Blended+Stable

	Stable Inpainting

	Ours