Shape cutouts

You can use a layer image with an opaque shape to either remove that shape from the image below that layer, leaving the shape to be transparent (e_cut_out), or conversely, use it like a cookie-cutter, to keep only that shape in the base image, and remove the rest (fl_cutter).

You can also use AI to keep or remove certain parts of an image (e_extract).

To embed a clipping path derived from a mask image into the metadata of a delivered JPEG or TIFF, use e_embed_clipping_path.

Remove a shape

The following example uses the cut_out effect to cut a logo shape (the overlay image) out of a base ruler image. A notebook photo underlay behind the cutout ruler lets you see the notebook paper and lines through the logo cutout:

Keep a shape

The following example uses the cutter flag to trim an image of a water drop based on the shape of a text layer (l_text:Unkempt_250_bold:Water/fl_cutter,fl_layer_apply). The text overlay is defined with the desired font and size of the resulting delivered image:

Blur the edges of an image

You can also use a semi-transparent PNG layer to create feathered edges. In this case, overlaying a square image with a black center and a white feathered border, using fl_relative to scale the overlay to the same width and height as the base image (c_scale,fl_relative,w_1.0,h_1.0), and applying fl_cutter to the overlay, keeps the opaque center of the mask and leaves the edge pixels partially transparent. To soften those edges against a non-transparent background, add a blurred underlay of the same image.

For example, using a feathered square PNG called square_feather:

Use AI to determine what to remove or keep in an image

You can use the e_extract transformation to specify what to remove or keep in the image using a natural language prompt.

For example, start with this image of a desk with picture frames:

You can extract the picture of the tree (e_extract:prompt_the%20picture%20of%20the%20tree):

Everything but the picture of the tree is now considered background, so you can then generate a new background for this picture, for example, an art gallery (e_gen_background_replace:prompt_an%20art%20gallery):

Or, you can invert the result of the extract transformation (invert_true), leaving everything but the picture of the tree:

And then generate a new picture in that space (e_gen_background_replace:prompt_a%20sketch%20of%20a%20tree):

To use a pre-determined background, you can use the extract effect in a layer. In this example, the multiple parameter is used to extract all the cameras in the image, and overlay them on a colorful background.

And this is the inverted result:

Using the extract effect in mask mode, you can achieve interesting results, for example, blend the mask overlay with the colorful image using e_overlay:

If your original image has transparency, you can keep the transparency in the output by using the preserve-alpha flag. For example, keep the transparent background when extracting the watch dial and inverting the result to keep only the strap:

Then you can achieve an interesting effect by using the extract effect again to keep only the dial and making it semi-transparent (e_extract:prompt_$prompt/o_30), then layering the first image, saved to a variable ($bracelet_current), on top of that. You can use a variable for the prompt too so you can easily reuse it ($prompt_!the%20watch%20dial!):

See full syntax: e_extract in the Transformation Reference.

Try it out: Content extraction.