Nano Banana custom block 🍌

Use the following code to use the block standalone:

from diffusers.modular_pipelines import ModularPipelineBlocks

banana_block = ModularPipelineBlocks.from_pretrained(
    "diffusers-internal-dev/nano-banana-modular",
    trust_remote_code=True,
)
banana = banana_block.init_pipeline()
output = banana(
    prompt="Create a picture of my cat eating a nano-banana in a fancy restaurant under the Gemini constellation"
)
print(f"{output.values['output_image'].size=}")
output.values["output_image"].save("generated_banana.png")

Result:

It accepts an image argument, too:

from diffusers.modular_pipelines import ModularPipelineBlocks
from diffusers.utils import load_image 

banana_block = ModularPipelineBlocks.from_pretrained(
    "diffusers-internal-dev/nano-banana-modular",
    trust_remote_code=True,
)
banana = banana_block.init_pipeline()
output = banana(
    prompt="Make Pikachu hold a sign that says 'Qwen Edit is awesome'",
    image=load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/yarn-art-pikachu.png")
)
print(f"{output.values['output_image'].size=}")
output.values["output_image"].save("edited_banana.png")

Result:

Original	Edited

Misc

Nano Banana: https://ai.google.dev/gemini-api/docs/image-generation