Nodes/☁️BizyAir Nodes/☁️BizyAir InpaintModelConditioning_v2
ComfyUI Node

☁️BizyAir InpaintModelConditioning_v2

BizyAir InpaintModelConditioning_v2 — ComfyUI Node

By siliconflow·Created 2 years ago·Updated 11 months ago· 855
☁️BizyAir InpaintModelConditioning_v2
  • positive
  • negative
  • vae
  • pixels
  • mask
  • positive
  • negative
  • latent
noise_maskfalse

What it is

Same job as the plain InpaintModelConditioning node, one extra dial. This is the node the community actually recommends reaching for when you're forcing real masked inpainting out of a modern editing model like Qwen-Image-Edit or Flux Kontext instead of relying on the model's default whole-frame re-render. The standing advice is specific: swap the empty latent for this conditioning node, get your denoise control back through a proper scheduler, and composite the result back over the original so nothing outside your mask ever survives a VAE round-trip. This v2 node is that same recipe with one more knob to fine-tune it.

How it works

It builds conditioning and a latent from your image and mask exactly like the base version - telling the model which pixels are fixed context and which region it's allowed to regenerate - but adds a noise_mask toggle that changes how the masked region gets seeded before sampling starts.

Inputs and outputs

  • positive / negative - BIZYAIR_CONDITIONING.
  • vae - BIZYAIR_VAE. Same rule as always with this pack: it only connects to other BizyAir-typed nodes, not a stock local VAE.
  • pixels - your source image.
  • mask - the editable region.
  • noise_mask (default false) - the node's own tooltip is blunt about this one: it adds a noise mask to the latent so sampling only happens within the mask, and "might improve results or completely break things depending on the model." That's not hedging for the sake of it - different base models genuinely respond to this differently, so it's worth toggling if your results look off.

Outputs: positive, negative, latent - same three-socket handoff into your BizyAir sampler as the non-v2 version.

How to install it

ComfyUI Manager: search "BizyAir," install, restart ComfyUI. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAir.git

A BizyAir API key is required before anything downstream of this node will actually execute - it's a cloud sampler on the other end, not a local one.

Common issues

noise_mask makes things worse, not better. Take the tooltip at face value - this isn't a "leave it on, it's strictly better" setting. If flipping it on degrades your result, that's the model telling you it doesn't like fully-noised sampling inside the mask; switch it back off and let the mask blend from the existing pixels instead.

Wiring errors between this and a non-BizyAir node. The BIZYAIR_CONDITIONING/BIZYAIR_VAE types are deliberately incompatible with the stock ComfyUI equivalents. If a connection refuses to snap, you've probably got a regular VAE loader or CLIP Text Encode somewhere in the chain where a BizyAir one needs to be.

Edges of the edit still look seamed. That's a compositing problem, not a conditioning problem - paste the sampled result back over the original using only the masked pixels (a crop-and-stitch node, or ImageCompositeMasked) rather than taking the fully decoded frame as your final output. Every VAE encode/decode cycle is slightly lossy, and skipping the composite step is the single most common reason people's "inpaints" quietly degrade the parts of the image they never meant to touch.

Category☁️BizyAir/conditioning/inpaint

Inputs (6)

NameTypeDefaultDescription
positiveBIZYAIR_CONDITIONING
negativeBIZYAIR_CONDITIONING
vaeBIZYAIR_VAE
pixelsIMAGE
maskMASK
noise_maskBOOLEANfalseAdd a noise mask to the latent so sampling will only happen within the mask. Might improve results or completely break things depending on the model.

Outputs (3)

NameTypeDescription
positiveBIZYAIR_CONDITIONING
negativeBIZYAIR_CONDITIONING
latentLATENT