Skip to content

Workflow examples

Chris Collins edited this page Nov 15, 2026 · 1 revision

Workflow examples

These wiring examples start with workflows that already generate successfully in your installation. Connect the finished IMAGE to CiviScribe. Model names, samplers, and node layouts depend on your chosen workflow.

Save a standard checkpoint image

Checkpoint -> text encoders -> positive and negative conditioning
           -> sampler model input
           -> VAE Decode VAE input

Empty latent -> sampler -> VAE Decode -> CiviScribe Images

Use PNG and portraits/%model_noext%_%counter:4% as the filename pattern. The save records the active checkpoint, prompts, available sampler settings, and final dimensions. Add a LoRA on the active model/CLIP path to record its filename, applicable strengths, and available identity evidence too.

Save five independent branches

Model A -> sampler A -> decode A -> CiviScribe A   branches/A/%counter:4%
Model B -> sampler B -> decode B -> CiviScribe B   branches/B/%counter:4%
Model C -> sampler C -> decode C -> CiviScribe C   branches/C/%counter:4%
Model D -> sampler D -> decode D -> CiviScribe D   branches/D/%counter:4%
Model E -> sampler E -> decode E -> CiviScribe E   branches/E/%counter:4%

Each CiviScribe node requests its own executing ID from ComfyUI and follows its own image branch. A loader elsewhere on the canvas cannot supply credits solely because it exists there. Give each branch a clear folder or prefix when you want independent image sequences.

Save PNG and JPEG from the same result

Final IMAGE -> CiviScribe PNG
           -> CiviScribe JPEG

Create two save nodes and connect the same final image to both. Select PNG on one and JPEG on the other. For example, use archive/%model_noext% for PNG and sharing/%model_noext% for JPEG. Choose a JPEG background if the image has alpha.

Each node writes its own file, encoding, metadata, counter, and optional sidecar. The preview for each node shows its actual saved result.

Record an image-to-image pass

Load Image -> VAE Encode -> sampler -> VAE Decode -> CiviScribe

Connect the final decode to CiviScribe. Known latent provenance and active sampling evidence can establish img2img classification and denoise. The final image dimensions describe the committed output, including later scaling.

Save an upscaled image

Generated IMAGE -> active upscaler -> CiviScribe

Connect after the upscaler to save the larger result and its final dimensions. Supported active upscaler resources can appear in the generation record. Use another save node before the upscale if you also want the earlier image.

Preserve an enhanced prompt

Enhancer -> final selected STRING -> text encoder -> generation
                                -> CiviScribe prompt override
Final IMAGE ------------------------------------> CiviScribe Images

Branch the last string after switches and trigger edits to the matching prompt override. For a subgraph, expose that final string as an output. Prompts and overrides

Test saving through the API without downloading a model

The repository includes a tiny API prompt that creates a synthetic 64×64 image and saves PNG and WebP copies with sidecars. It exercises saving and counter placement. Because it contains no model or sampler, generation settings and resource identities appropriately remain unknown.

This is a ComfyUI API request body for POST /prompt. API users can submit it to their own running instance. It is separate from the editable UI workflow format. The second example contains a standard checkpoint workflow with REPLACE_WITH_YOUR_CHECKPOINT.safetensors for you to replace with an installed checkpoint's exact selector. Synthetic save example · Checkpoint example

Clone this wiki locally