Reference Images Pixaroma

How to use it

• Drop pictures on the node, click Add, or select the node and paste one with Ctrl+V.
• Under each picture, pick what it is for: Character, Outfit, Pose, Style, Background and more. The small box under it can name it, like the woman in red.
• Write what to make in Your prompt. Write @1 for picture 1, @2 for picture 2 (or click a picture's number to add it), like Put @1 in the outfit from @2 on a sunny beach.
• Wire prompt into the prompt of your model's text encode, and the outputs 1, 2... on the right edge (picture 1, picture 2) into its picture inputs, in the same order. One output can feed several nodes.
• Run. Prompt sent shows the exact text the model gets.

Why the tags matter

Every model names its reference pictures its own way, and a prompt that says it the wrong way leaves the model guessing which picture is which. This node writes them for you:

Qwen Image 2.1

<image1>, <image2>...

FLUX 3 Image Klein

<ref_image_0>, <ref_image_1>... It counts from 0, so picture 1 is <ref_image_0>.

MiniMax H3

<Picture 1>, <Picture 2>...

Plain words

image 1, image 2, for Flux 2 Klein and most other models.

Custom

Your own pattern, set in the gear. {n} is the number from 1, {i} from 0.

Model: Auto

Auto follows the prompt wire to see where it goes: into Text Encode Qwen Image 2.1 it writes Qwen's tags, into a FLUX 3 reference node FLUX 3's, into a MiniMax H3 node H3's, and plain words otherwise. The button shows what it found. When the prompt goes through other nodes first (Prompt, Find and Replace, Text Join...), Auto still follows it. Pick the model yourself when it can't tell.

The picture list

Before your own words the node adds one sentence that says what each picture is for, like Use the character from <image1>, the outfit from <image2> and the background from <image3>. It follows the roles and names you set. In the gear you can move it after your words, turn it off, or change the words used for each role.

The buttons on a picture

The number

Click it to write @1 (or @2...) where the cursor is in your prompt.

Click the picture

Use another picture in its place. Dropping a picture on it does the same.

◀ ▶

Move it left or right. Its number changes, and so does which output it comes out of. The @ numbers in your prompt move with it (a prompt that comes over the text wire keeps its numbers: check them).

×

Remove it. Its output goes with it, and the wires and the @ numbers of the pictures after it stay with their pictures. Its own @ number in your prompt becomes its words (like the outfit). A prompt that comes over the text wire keeps its numbers: check them.

What comes out

prompt

Your words with every @1, @2 turned into the model's tag, plus the picture list.

1, 2, 3...

The pictures, one output each (image_1, image_2... in menus), in the order shown. A new dot appears on the right edge for every picture you add, up to 12.

Settings (the gear)

The picture list

Before your prompt, after it, or off.

Transparent parts

The colour put behind a picture with a see-through background: white, grey or black.

Picture size

Keep them as they are, or make big ones smaller (up to 1 or 2 megapixels). Smaller is faster.

Round the size to

Trims a few pixels off the edges so the size divides by 16, 32 or 64. Auto trims to 32 for Qwen Image 2.1, whose text encode would otherwise stretch each picture a little.

The words for each role

How each role is said in the picture list.

Button colour

The accent colour of this node.

Tips

• Give each picture one job. A picture of a person in a dress used for the outfit can bring the person along too: say it in your words (only the dress from @2).
• With Qwen Image 2.1 and FLUX 3, the result takes the shape of picture 1. Put the picture whose shape you want first, or set the size with an Empty Latent.
• Want your prompt from another node? Wire it into text. The box on the node then shows it is wired, and @1, @2 in the wired text still become the tags.
• The pictures are saved in ComfyUI's input folder. A workflow shared with someone else needs the pictures too.
• Drag the node wider for bigger pictures (up to 4 a row), or taller for a bigger Your prompt box. It never gets small enough to hide a picture.
• Only need pictures, no prompt? Leave prompt unwired and use the outputs alone, for example one picture per first and last frame of a video.

Text too small?

Right-click the node > Text size.

All text boxes: Bigger / Smaller / Reset changes ComfyUI's own setting (Settings > Appearance > Node Widget > Textarea widget font size). Every text box grows together, ComfyUI's own boxes too, in every workflow.

Only this node: Bigger / Smaller / Reset changes this node alone. It is saved in the workflow, so it travels with it.

Spell check

Turn it on in Settings > Comfy > Node Widget > Textarea widget spellcheck (or search the settings for spellcheck). It is off by default.

The text boxes of this node follow that setting, like ComfyUI's own boxes: a word the browser does not know gets a red wavy underline.