model-and-prompt-from-metadata

Model and Prompt from Metadata

Language: 日本語 | 中文

---

A ComfyUI custom node designed to quickly reuse metadata from images generated with SD1.5 / SDXL / Illustrious models, accelerating your overall workflow.

Simply drop a PNG or WebP image, or a workflow JSON, onto the node to extract the checkpoint, VAE, and prompts from the embedded metadata and apply them automatically. Instantly reproduce past generation settings and shorten your trial-and-error cycle.

Also supports images generated with Stable Diffusion WebUI / SD Forge neo / Fooocus, and ComfyUI-Custom-Scripts Workflow Images.

Out-of-Scope Models

UNet-based models such as Flux, QWEN, and zImage use a different architecture from checkpoint-based models and are outside the scope of this node. Dropping a file that contains these models will display the detected model names.

For workflows using UNet-based models, Workflow Studio is recommended. Its Library → Information tab fully supports UNet-based architectures (Flux.2, Qwen-Image, Z-Image, WAN2.2, HiDream, and more), and lets you drag detected models and prompts directly onto the ComfyUI canvas. See Workflow Studio Integration below for details.

UI language: Automatically switches between English / Japanese / Chinese based on your browser's language settings.

---

Screenshot

All four nodes in default state

All four nodes: LoRA from Metadata, Model-Prompt from Metadata, Model from Metadata, and CLIP Text Encode edit+.

Nodes with metadata loaded

Left: LoRA from Metadata — LoRA detected and auto-assigned to slot 1. Center: Model-Prompt from Metadata — checkpoint, VAE, and prompt auto-selected. Right: Model from Metadata — UNet-based model (Flux) detected, prompting use of the workflow directly.

---

Nodes

Model from Metadata (ImageMetadataCheckpointLoader)

Category: loaders

Drop a PNG, WebP, or JSON to load the checkpoint and VAE.

Outputs

Name	Type	Description
model	MODEL	Loaded model
clip	CLIP	CLIP
vae	VAE	VAE (uses checkpoint's built-in VAE when "None" is selected)
image	IMAGE	The dropped image (a 64x64 black image when nothing is dropped)

---

Model-Prompt from Metadata (ImageMetadataPromptLoader)

Category: loaders

In addition to the checkpoint and VAE, also extracts and encodes positive/negative prompts. Accepts PNG, WebP, and JSON.

Outputs

Name	Type	Description
model	MODEL	Loaded model
clip	CLIP	CLIP
vae	VAE	VAE
positive	CONDITIONING	Positive conditioning
negative	CONDITIONING	Negative conditioning
positive_text	STRING	Positive prompt (raw text)
negative_text	STRING	Negative prompt (raw text)
image	IMAGE	The dropped image (a 64x64 black image when nothing is dropped)

---

LoRA from Metadata (ImageMetadataLoRALoader)

Category: loaders

Applies up to 3 LoRA models sequentially. Drop a PNG, WebP, or JSON to auto-detect and assign LoRAs from metadata.

Inputs

Name	Type	Description
model	MODEL	Model
clip	CLIP	CLIP
metadata	METADATA	(Optional) Metadata from an upstream node

Outputs

Name	Type	Description
model	MODEL	Model with LoRAs applied
clip	CLIP	CLIP with LoRAs applied

---

CLIP Text Encode edit+ (CLIPTextEncodeEditPlus)

Category: conditioning

A CLIP encoder with four modes: use the raw received text (RAW), a manually edited version (EDIT), or insert a second text before (front) or after (back) the main prompt. Use two instances — one for positive, one for negative.

• EDIT textarea: Pre-filled with the received text on first connection; freely editable
• **RAW / EDIT / front / bac