Cheat Sheet

ComfyUI Metadata Field Reference

Every metadata field ComfyUI embeds — what it is, where it comes from, and what survives editing

numonic.ai

AI-Native Asset Management

PNG Chunks (Raw Data)

ComfyUI embeds two text chunks in every PNG it saves. These are the root of all extractable metadata.

Field	Type	Produced By	Description
`prompt`	JSON string	`All workflows`	Execution-format workflow: node map with class_type and inputs for every node. This is the machine-readable version.
`workflow`	JSON string	`All workflows`	UI-format workflow: node positions, links array, visual layout. This is what you see in the ComfyUI editor.

Generation Parameters

Core generation settings extracted from sampler and latent image nodes in the workflow.

Field	Type	Produced By	Description
`steps`	number	`KSamplerKSamplerAdvanced`	Number of denoising steps.
`cfg`	number	`KSamplerKSamplerAdvanced`	Classifier-free guidance scale. Higher values follow the prompt more closely.
`sampler`	string	`KSamplerKSamplerAdvanced`	Sampler algorithm name (e.g., euler, dpmpp_2m, uni_pc).
`scheduler`	string	`KSamplerKSamplerAdvanced`	Noise scheduler (e.g., normal, karras, exponential, sgm_uniform).
`seed`	number	`KSamplerKSamplerAdvanced`	Random seed for reproducibility. Same seed + same settings = same image.
`denoise`	number	`KSamplerKSamplerAdvanced`	Denoising strength (0.0–1.0). Lower values preserve more of the input image.
`width`	number	`EmptyLatentImageEmptySD3LatentImage`	Output image width in pixels.
`height`	number	`EmptyLatentImageEmptySD3LatentImage`	Output image height in pixels.
`batchSize`	number	`EmptyLatentImageEmptySD3LatentImage`	Number of images generated per batch.
`model`	string	`CheckpointLoaderSimpleCheckpointLoader`	Checkpoint filename (e.g., sd_xl_base_1.0.safetensors). Filename only, not the model contents.

Prompt Text

Positive and negative text prompts extracted from CLIP encoding nodes.

Field	Type	Produced By	Description
`prompt`	string	`CLIPTextEncode`	Positive prompt text. Extracted from the first non-negative CLIPTextEncode node found.
`negativePrompt`	string	`CLIPTextEncode`	Negative prompt text. Identified by node title containing "negative" or "neg".

Model References

All model files referenced in the workflow — checkpoints, LoRAs, VAEs, and ControlNets.

Field	Type	Produced By	Description
`models[].type`	string enum	`All loader nodes`	Model type: checkpoint, lora, vae, controlnet, or other. Inferred from the loader node type.
`models[].name`	string	`CheckpointLoaderSimpleLoraLoaderVAELoaderControlNetLoader`	Model filename from the loader node input (ckpt_name, lora_name, vae_name, or control_net_name).
`models[].hash`	string \| null	`Optional`	Model file hash for deduplication. Not always populated — depends on the workflow configuration.
`models[].config`	object \| null	`LoraLoaderCustom nodes`	Additional configuration: LoRA strength_model/strength_clip, custom node parameters.

Video Parameters

Fields extracted from MP4 files created by video output nodes (ComfyUI-VideoHelperSuite, AnimateDiff).

Field	Type	Produced By	Description
`frameRate`	number	`VHS_VideoCombineAnimateDiffCombine`	Output video frame rate (fps).
`videoFormat`	string	`VHS_VideoCombine`	Video container format (e.g., video/h264-mp4).
`crf`	number	`VHS_VideoCombine`	Constant Rate Factor — quality setting where lower = better quality, higher = smaller file.
`pixelFormat`	string	`VHS_VideoCombine`	Pixel format (e.g., yuv420p, yuv444p).
`loopCount`	number	`VHS_VideoCombine`	Number of times the video loops (0 = no loop).
`pingpong`	boolean	`VHS_VideoCombine`	Whether the video plays forward then backward for seamless looping.
`width / height`	number	`MP4 container`	Video dimensions in pixels (from container metadata).
`duration`	number	`MP4 container`	Video duration in seconds.
`bitRate`	number	`MP4 container`	Video bitrate in bits per second.
`codec`	string	`MP4 container`	Video codec (e.g., h264, h265).
`encoder`	string	`MP4 container`	Encoder software name (from the MP4 ©too atom).

What Survives? Metadata Survival Matrix

ComfyUI metadata lives inside PNG text chunks. Any operation that strips or re-encodes the image can destroy it.

Metadata	Original PNG	PNG Re-save	PNG→JPEG/WebP	Social Upload	Numonic (stored)	Numonic Export
Workflow JSON	Kept	Kept	Lost	Lost	Kept	Depends on preset
Prompt data	Kept	Kept	Lost	Lost	Kept	Depends on preset
Generation params	Kept	Kept	Lost	Lost	Kept	Depends on preset
Model names	Kept	Kept	Lost	Lost	Kept	Depends on preset
Seed values	Kept	Kept	Lost	Lost	Kept	Depends on preset
EXIF technical	Kept	Kept	Kept	Depends on preset	Kept	Kept
Artist / Copyright	Kept	Kept	Kept	Lost	Kept	Depends on preset

KeptDepends on presetLost

Numonic never loses your metadata. The Stored column shows what Numonic keeps internally — everything, always. The Export column shows what ends up in the exported file, which you control via presets: Archive keeps everything, Portfolio keeps attribution but strips workflows, and Share strips sensitive data for social media. Your originals and metadata are never touched.

The Golden Rule

Always keep your original PNGs. ComfyUI embeds the full workflow, prompt, and generation parameters directly in the PNG file. Converting to JPEG, WebP, or any other format permanently destroys this metadata. Treat your original PNGs as "digital negatives" — copy them for sharing, never move or convert the originals.

Better yet, automate it. Numonic reads every field in this reference on import, stores it in a searchable database, and lets you export with full control over what metadata to include. Your originals stay untouched.

Stop losing your ComfyUI metadata

Every field on this page is automatically extracted, searchable, and preserved when you import to Numonic. Search by seed, filter by model, export with the metadata you choose.

Try Numonic FreeFree tier included · No credit card required

From The Complete Guide to ComfyUI Asset Management — numonic.ai/guides/comfyui-asset-management

Free to share and print. Try Numonic free · Attribution appreciated.