☁️BizyAir ControlNetInpaintingAliMamaApply

This is the cloud wrapper for a specific, well-known tool: Alimama's ControlNet-Inpainting-Beta, one of the two standard choices (alongside FLUX Fill) for doing serious inpainting on Flux-class models. It's the same DiT-era ControlNet apply node as its SD3 sibling, plus a mask input, which turns it from "apply structural conditioning" into "apply structural conditioning specifically to a masked region while regenerating it."

Why a dedicated inpainting ControlNet beats the basic approach. There are two ways to inpaint. The simple version - the one that doesn't need a specialized model at all - is regenerating the whole image and letting the UI composite the new content onto the original inside your mask boundary; it's easy but prone to visible seams where the regenerated patch meets the untouched original, since the model has no idea a boundary even exists. A dedicated mask-aware inpainting ControlNet like this one is a genuinely different mechanism: the model is trained knowing where the mask boundary is, so it blends the transition itself instead of leaving that job to post-hoc compositing. That's the entire value proposition here, and it's why Alimama's model specifically became the standard pick for Flux inpainting rather than people just using a generic ControlNet with a mask bolted on.

Inputs that matter. positive and negative (BIZYAIR_CONDITIONING), control_net (BIZYAIR_CONTROL_NET - load Alimama's inpainting model, or whatever compatible inpainting ControlNet BizyAir's catalog offers, through a BizyAir ControlNet loader), vae (BIZYAIR_VAE, required for the same reason it's required on the plain SD3-style apply node - DiT ControlNets need the reference image encoded into latent space first), image (the source image you're inpainting), and mask (a plain, un-prefixed MASK - the region to regenerate). strength (default 1, range 0–10, step 0.01) and the start_percent/end_percent step window (0–1, step 0.001) behave the same as on every other ControlNet apply node in this pack - dial the end percent down if you want the model reclaiming fine detail in the later steps rather than staying pinned to the ControlNet's guidance the whole way through.

Outputs: positive and negative, both BIZYAIR_CONDITIONING - feed these into your sampler/guider in place of your original conditioning pair, same pattern as the rest of the ControlNet apply family here.

Install: search "BizyAir" in ComfyUI Manager and click install, cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git and restart, or comfy node install bizyair - one pack install covers this node along with the rest of BizyAir's ControlNet lineup. It's a cloud-executed node, so it needs a registered BizyAir account with your API key set before it'll run, same as everything else here.

Troubleshooting. Make sure your mask actually covers the region you intend to regenerate and not its inverse - an accidentally-inverted mask from an upstream node is the single most common reason an inpaint "does nothing" or changes everything except the intended area. If the blend at the mask boundary still looks seamy despite using a dedicated inpainting ControlNet, try raising strength slightly or narrowing start_percent/end_percent so the ControlNet's guidance is concentrated where it matters most. As with any node in this pack, a stall or unclear failure is worth an API-key check before assuming the mask or ControlNet setup is the problem.