ComfyUI-SaveIntermediates

ComfyUI-SaveIntermediates

Save and stream intermediate sampling steps during diffusion. Perfect for showing generation progress on your frontend instead of a loading spinner.

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/DanielBartolic/ComfyUI-SaveIntermediates.git

Usage

For Serverless/API Streaming

Connect the node between your model loader and sampler:

Load Checkpoint → [Save Intermediate Steps] → ClownSampler → ...
                         ↓
                   (outputs to progress folder)

Your frontend can poll progress.json to get:

• Current step / total steps
• Progress percentage
• Path to latest image
• Optional base64 encoded image

Workflow with Image Output

Load Checkpoint → [Save Intermediate Steps] → ClownSampler → [Get Intermediate Images] → Preview/Save
                                                    ↓
                                              (latent output)

Nodes

Save Intermediate Steps

Wraps model and saves TAESD previews at each step.

Inputs:

• model - From your model loader
• steps - Total sampling steps (for progress calculation)
• job_id - Unique ID for this job (creates subfolder)
• output_folder - Base folder name (default: "progress")
• filename_prefix - Prefix for files (default: "step")
• format - jpeg or png
• include_base64 - Include base64 in progress.json

Output:

• model - Pass to your sampler

Save Intermediate Steps (Advanced)

Same as above with filtering options:

• save_every_n - Save every Nth step
• start_step / end_step - Range of steps to save

Get Intermediate Images

Collects all saved intermediates as IMAGE batch after sampling.

Inputs:

• latent - Connect from sampler (ensures this runs after)
• job_id - Match the job_id used in Save node
• output_folder - Match the folder used

Output:

• images - Batch of all intermediate images

Output Structure

ComfyUI/output/progress/{job_id}/
├── progress.json      # Poll this for status
├── latest.jpeg        # Always the most recent step
├── step_0000.jpeg
├── step_0001.jpeg
├── step_0002.jpeg
...

progress.json Format

{
  "job_id": "abc123",
  "status": "generating",
  "current_step": 15,
  "total_steps": 20,
  "progress_percent": 75,
  "latest_image": "step_0015.jpeg",
  "timestamp": 1704567890.123,
  "image_base64": "..."  // if include_base64 enabled
}

Frontend Integration Example

// Poll for progress
async function pollProgress(jobId) {
  const response = await fetch(`/output/progress/${jobId}/progress.json`);
  const data = await response.json();

  if (data.status === 'generating') {
    // Show preview image
    const imgUrl = `/output/progress/${jobId}/${data.latest_image}`;
    updatePreview(imgUrl);
    updateProgressBar(data.progress_percent);
  }

  if (data.status !== 'completed') {
    setTimeout(() => pollProgress(jobId), 500);
  }
}

License

MIT