For ControlNetInpaintingAliMamaApply, do not infer the mask direction from a general black-and-white convention. The documented special case is conditional: when the selected ControlNet has concat_mask enabled, the node inverts the input mask, applies that inverted mask to the image, and includes the inverted mask in extra concatenation data sent to ControlNet. Check that setting and those three data roles before deciding which region the control input represents. The documentation does not establish a universal rule for masks in other nodes or workflows.
What the node documents
The node takes positive and negative conditioning, a ControlNet model, a VAE, an image, a mask, and strength and timing controls. Its documentation describes the mask as defining the areas to be inpainted, then adds a more specific note for ControlNets with concat_mask enabled: the mask is inverted, and the documentation says the inverted mask is applied to the image before processing and included in extra data passed to the ControlNet.
Keep those facts separate. The input mask is what you connect to the node. The inverted mask is the transformed mask in the documented concat_mask path. The extra concatenation data is what the documentation says accompanies the control input; it is not a claim that the node simply forwards the original mask unchanged. The documentation does not specify a universal white/black convention, how soft or gray values are handled, or the visual result of a particular model.
Trace the mask before you judge the region
Use this ordered check on the workflow you are about to run:
- Confirm the exact node is
ControlNetInpaintingAliMamaApply, rather than a different inpainting or ControlNet node. - Identify the
control_netconnected to it and determine whether that ControlNet hasconcat_maskenabled. Do not assume the setting from the node’s name. - Mark the region represented by the connected input
maskin your own notes. Describe it spatially, for example, “the rectangle over the sign,” rather than calling it simply the editable or protected region. - If
concat_maskis enabled, account for the documented inversion: the mask applied to the image is inverted relative to the input mask, and the inverted mask is also included in the extra concatenation data sent to ControlNet. - Inspect the
imageconnected to the same node and check that the mask you are reasoning about aligns with the intended area in that image. The documentation describes these as paired node inputs; it does not promise a preview of the transformed data. - Only then assess the workflow’s expected control input. Keep
strength,start_percent, andend_percentas separate settings: the docs describe them as influence controls, not as ways to change the mask’s inversion rule.
This check is about the mask-processing path, not a guarantee that only one image region will change in the final generation.
A hypothetical rectangle check
Suppose, hypothetically, the input image is 512 × 512 pixels and the input mask marks a 128 × 96 rectangle at the upper left as white, with the rest black. Treat these values and colors only as a labeled example; they are not a claim about a universal mask convention or measured output. If the connected ControlNet has concat_mask enabled, the documented operation is to invert that input mask, apply the inverted mask to the image, and include the inverted mask in the extra concatenation data.
| Checkpoint | Hypothetical observation | What follows from the documentation |
|---|---|---|
| Node and setting | ControlNetInpaintingAliMamaApply; concat_mask enabled | The special mask-processing note applies |
| Input mask | White rectangle at upper left; black elsewhere | This describes the connected input, not the post-inversion mask |
| After inversion | The rectangle and surrounding values are reversed | The inverted mask is the one applied to the image and included in extra data |
| Intended control area | Creator expects the upper-left rectangle to guide control | Verify the intended meaning against the transformed path; do not conclude final pixels changed only there |
The table is a reasoning aid, not a preview or result from running ComfyUI. If you cannot tell which ControlNet is connected or whether its setting is enabled, stop short of asserting which region the model will treat as active.
If the expected region still seems wrong
First resolve uncertainty at the input boundary: confirm the target node, connected ControlNet, concat_mask state, and the region represented by the mask you connected. Then inspect how the mask and image are produced upstream, including whether a transform changed their alignment. The supplied node documentation supports checking the image and mask inputs, but it does not specify resizing, cropping, thresholding, edge treatment, or a method for previewing the internal inverted tensors. Do not invent a diagnosis for those details from the inversion note alone.
If the setting is disabled, the cited special-case note does not describe an inversion-and-concatenation operation for that condition. It does not, by itself, fully specify every other internal behavior. If the setting is enabled and the workflow still does not match your expectation, consult documentation for the specific ControlNet or inspect the relevant implementation before attributing a visible result to the mask. Changing strength or timing may alter ControlNet influence, but the cited documentation gives no basis for saying those changes will correct a direction mismatch or preserve a region exactly.
Common questions
Does white always mean “inpaint this area” here?
The node documentation labels the mask as defining inpaint areas, but it does not provide a universal black/white mapping. In the documented concat_mask case, the input is inverted before the described image application and extra-data concatenation. Record the actual input convention for your workflow rather than importing one from another tool.
Is the original mask sent to the ControlNet unchanged?
The special-case note says the inverted mask is included in the extra concatenation data. It does not say that the original mask is included unchanged in that data.
Does this tell me exactly which final pixels will change?
No. It describes conditioning and mask handling, not a pixel-level output guarantee. It also does not establish that all other ControlNet nodes use this behavior. For this decision, use the official ControlNetInpaintingAliMamaApply node documentation as the scope; investigate model-specific behavior separately.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.