ComfyUI Node

AIO PID Info

Grab the latent mid-sampling, sigma and step included

By helto4real·Created 3 months ago·Updated 23 days ago· 1
AIO PID Info
  • pid_info
  • latent
  • sigma
  • step

Sampling doesn't have to be an all-or-nothing black box. AIO Image Generate has a pid_capture_step input - set a sampler step, and the pipeline captures the latent at that step plus its sigma, bundled into the pid_info output. AIOPIDInfo unpacks that bundle into plain LATENT, FLOAT (sigma), and INT (step) values you can wire anywhere. It's the same class of utility node as AIOModelInfo and AIOInpaintInfo: one opaque bundle in, typed sockets out.

What it gives you

One required input, pid_info, from AIO Image Generate. Three outputs:

  • latent - the captured latent at the chosen sampler step.
  • sigma - the noise level at capture time.
  • step - which step it actually captured (useful when you left pid_capture_step at 0 and want to know what "auto near the end" resolved to).

On the main node, pid_capture_step defaults to 0, which auto-selects a step near the end of sampling. Set it explicitly to a mid-sampling step to catch the latent at a particular noise level.

When you'd actually use it

This is a specialist node, and honest about being one. The practical uses:

  • Mid-sampling inspection - decode the captured latent to see how the image looked partway through, useful for diagnosing why a run converges badly.
  • Detail / refinement branches - feed the captured latent (at a mid or late sigma) into a secondary sampler or an upscale-detail path, effectively branching off a partially-denoised image instead of starting from the final result.
  • Sigma-aware logic - wire sigma into a node that gates behavior on noise level, or log it with your runs.

If none of those sound like you, you don't need this node. The main node's image, latent and run_info outputs cover normal generation; PID capture is for people doing non-standard things with the middle of a denoising schedule.

Install and wiring

Ships with the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/helto4real/comfyui-all-on-one-image-generation-node
cd comfyui-all-on-one-image-generation-node
python -m pip install -r requirements.txt

Or ComfyUI Manager → search "AIO Image Generate" → install → restart; it's under AIO/Image. Connect AIO Image Generate.pid_infoAIOPIDInfo.pid_info, set pid_capture_step on the main node, and wire the latent/sigma/step outputs where they need to go.

Gotchas

  • The capture only happens when it's wanted. Like the other info bundles, the heavy values are only produced when the matching output is reachable - so connect the outputs you're using or the capture may resolve to a no-op.
  • step 0 means auto. If you want a specific moment in the schedule, set pid_capture_step to something explicit; 0's "near the end" is a guess, and the step output is there to tell you what it picked.

A small node for a niche job: peeking into the middle of a denoising run and taking a snapshot. When that's your job, this is exactly the tool.

CategoryAIO/Image

Inputs (1)

NameTypeDefaultDescription
pid_infoAIO_PID_INFOPID capture values from AIO Image Generate.

Outputs (3)

NameTypeDescription
latentLATENT
sigmaFLOAT
stepINT