Image

The image tools let an assistant create and change images in Nodaro: generate from a prompt, transform or edit an existing image, cut a mask, composite several images or layers, and describe what an image shows. Each tool except suggest_overlay_placement, save_image_defaults and the palette and moodboard tools starts a job and returns its id, and the result is saved to your library. Every image tool except the palette and moodboard tools needs the workflows:execute permission.

generate_image

Creates an image from a text prompt with one of Nodaro's image models. It is the tool behind requests such as "make an image of…", and it works like the Generate Image node.

Permission: workflows:execute. Credits: the price of the model at the chosen resolution or quality.

Returns: a job id. In clients with MCP Apps, the card shows the image when it is ready.

Several images in one call

• count makes up to 4 images of the same request. Each image is its own job.
• models runs the same request on 2 to 4 models. With both, count is per model, so one call makes at most 8 images.
• The answer lists every job id, grouped by model, with the batch id. The card and jobId follow the first job.
• Each model gets its own settings. Levers such as aspect ratio and resolution are fitted to each model, as on a single call.
• The batch is charged as a whole. Nodaro checks your balance for the whole batch and creates all of its jobs, or none. An error answer means nothing was charged.
• Refusals. A count outside 1 to 4, more than 4 models, model and models together, or a list that names a model that is not an image model is refused. The error names the problem. Model ids in models match without regard to capital letters.

Apply a preset

presetId applies a built-in or saved preset exactly as the editor applies it:

• The preset's prompt and its text before and after the prompt are used. Your own prompt is wrapped in that text.
• Placeholders written as {name || default} are filled with their defaults.
• The preset's style and negative prompt are passed the way the model takes them.
• Any setting the preset leaves unset gets the editor's default. Your saved image defaults do not apply to a preset call.

A placeholder with no default, such as {BRAND}, needs a value in variables. Without one, the tool names the missing value. With count, the tool makes that many images of the preset's request. With models, it applies the preset once for each model. You cannot combine presetId with palette_id or moodboard_id yet: the call is refused.

Palettes and moodboards

palette_id names every color of a palette in the prompt, by an everyday name and its hex code. moodboard_id adds the moodboard's description to the prompt and sends its images as style references, after your own references, up to the model's limit. The answer says when some images did not fit. Create and find palettes and moodboards with the tools below.

Output format

generate_image, modify_image and image_to_image take output_format on GPT Image 2.6 Flare and Sunburst. list_models lists the formats of each model as outputFormats.

• Formats. png (default), jpeg, webp, gif or svg. jpg is read as jpeg.
• GIF. An animated square GIF of 512 × 512 pixels, with 2 to 16 frames. gif_frames sets the number of frames, default 9. gif_frame_duration_ms sets how long each frame shows, from 20 to 2,000 ms, default 200. A GIF ignores the aspect ratio and the resolution.
• SVG. svg_level sets the detail: standard, detailed or max.
• Price. PNG, JPEG and WebP cost the same. A GIF is priced by its quality and its number of frames. An SVG costs the PNG price plus a part for its level. See Formats on the Generate Image page.
• Limits. A transparent background is kept only in PNG and WebP. A mask with a GIF or an SVG is refused.

What the server changed is listed in the answer's adjustments.

modify_image

Edits, restyles or transforms an existing image with a prompt: change a color, swap a background, apply a style, extend or repaint part of the frame. It works like the Modify Image node.

Permission: workflows:execute. Credits: the price of the model.

Returns: a job id.

image_to_image

Transforms an image with a prompt, with options for masks and for composing several references in one image.

Permission: workflows:execute. Credits: the price of the model.

Your saved image defaults fill in what the call leaves out. A value the call names always wins.

• Aspect ratio, resolution and quality. Each saved value is fitted to the model. A saved aspect_ratio: "auto" works like a named ratio. A saved resolution or quality that the model does not take is left out.
• Model. The saved model is used only when an edit can run it: an edit model, or a model that generates and edits under one id, such as nano-banana-pro. A text-to-image model such as gpt-image-2 is not used for an edit.
• With neither. The GPT Image 2, 2.5 and 2.6 edit models and Seedream 5.5 run at auto and keep the photo's shape. GPT Image 2, GPT Image 2.5 and Seedream 5.5 receive their own auto, and GPT Image 2.6 the supported ratio closest to the photo. Any other model runs at its own default ratio.

Returns: a job id.

Masks

mask_url is a black and white image: white is the area to change, and black is kept. image_to_image and modify_image take it the same way. On GPT Image 2.6, the rest of the picture is kept exactly as it was: the source is pasted back everywhere outside the mask, and the result has the source's size. The run keeps the photo's shape and renders at the size (1K, 2K or 4K) nearest the photo, and is priced at that size. The answer's adjustments list each change. A mask costs nothing extra and does not count as an input image. Make a mask with generate_mask.

edit_image

Upscales an image, removes its background, or applies a targeted edit. Choose the operation with model.

Permission: workflows:execute. Credits: the price of the operation, in the table below.

To edit only some regions of a Grok Imagine 2 image, work in two steps. First run grok-2-segment with the image's task_id, and read the named regions from the job's segments. Then run grok-2-edit with the same task_id, a prompt, and the mask_indexes of the regions to change. The region map is free.

grok-upscale is no longer offered: it never upscaled an image. A call that still names it is refused with a message that names the replacement, recraft-upscale or topaz-image-upscale.

Returns: a job id.

split_into_layers

Splits one image into a base picture and separate layers, each element on its own transparent PNG. Each layer comes with its name, description, position and stacking order. It works like the Split into Layers node, and is not listed where that node is not offered.

Permission: workflows:execute. Credits: charged per image returned, the base included. A run holds the price of max_layers + 1 images at its size, and is charged only for the images it returns. 1K and 1.5K are priced alike, and so are 2K and auto.

Returns: a job id. The job's output holds:

• imageUrl: the base picture, an opaque PNG.
• layers: each layer's file, from bottom to top, and layerCount.
• layersNotSaved: the layers that could not be saved. They are not charged.
• layerData: one entry per layer, in the same order, with name, description, zIndex, width, height, box, boxNormalized, url and baseUrl.

zIndex is the stacking order: 1 sits on the base, and higher is drawn above. box is where the layer sits on the base, in the base's pixels, and boxNormalized is the same box on a 0 to 1000 scale. A layer file is cropped to its element, so scale it to its box to put it back in place. A run that finds no element returns the base only, at the price of one image.

generate_mask

Makes a black and white mask of what you describe, for example "the person" or "sky and clouds". White is the selected area. Use the mask with image_to_image, edit_image or the inpainting fields of generate_image. It works like the Generate Mask node.

Permission: workflows:execute. Credits: 55.

Returns: a job id. The result is a PNG mask.

image_collage

Puts 2 to 30 images into one large image. No image is ever cropped: the smart layout arranges them in rows at their own shapes, and the grid layout uses equal cells with the empty space filled by the background color. It works like the Image Collage node, and suits storyboards and contact sheets.

Permission: workflows:execute. Credits: 22 at 2K, 44 at 4K.

Returns: a job id.

image_overlay

Places 1 to 12 layers on a base image, pixel-exactly: logos, cut-outs, text, QR codes and shapes. No AI model runs, so the result is the same every time. It works like the Image Overlay node, which documents every option.

Permission: workflows:execute. Credits: 11, and a little more for each extra platform size in variants, as on the Image Overlay node.

The main layer fields:

A watermark in the bottom right corner is anchor: "bottom-right", x: -4, y: -6, width: 12. SVG logos are rendered sharp at the target size.

Returns: a job id. The result carries the image, maskUrl, its width and height, and variants, a list of { id, label, width, height, url }.

suggest_overlay_placement

Asks a vision model where one layer should sit on a base image. The model looks at the picture and keeps the layer off faces, the subject and busy texture. Nothing is composited: pass the answer to image_overlay.

Permission: workflows:execute. Credits: one image description, charged like image_to_text.

Returns: anchor, x, y and width in the same percent units as image_overlay, and a one-sentence reason.

image_to_text

Describes an image, or answers a question about it, with a vision model. Use it for captions, alt text, scene analysis or a specific question, such as "list all the text in the image". It works like the Describe Image node.

Permission: workflows:execute. Credits: one LLM call, charged for the tokens it actually used at the rate of the model that ran it. A hold is reserved before the call, and a balance below the hold is refused with 402. A longer description costs more. The editor's Run button shows ≈N, an estimate of a typical call. See How LLM calls are charged.

Returns: a job id. The text is in the job's output, and the card shows it with a Copy button.

save_image_defaults

Saves your preferred image settings on your account, as the defaults of later generate_image, modify_image and image_to_image calls. Only the fields you pass change. In clients with MCP Apps, the Save as default chip on an image card calls this tool.

• A value the call names wins. A saved value fills in what the call leaves out. With neither, the tool's own default applies. A saved aspect_ratio: "auto" works like a named ratio.
• One saved model for all three tools. The star on an image card saves the model that made the picture. An edit (modify_image, image_to_image) uses the saved model only when an edit can run it. generate_image uses it only when the model generates images. So after a star on an edit's result, generate_image runs its own default.
• Presets. A call with presetId does not read your saved defaults.

Permission: workflows:execute. Credits: free.

Pass null for a field to clear it.

Returns: a confirmation.

list_palettes

Lists your color palettes, newest first: the name, each color by its everyday name and hex code, the note and the id. Pass an id to generate_image as palette_id.

Permission: assets:read. Credits: free.

Returns: a page of palettes, the same list as GET /v1/palettes.

save_palette

Creates a palette, or changes one. To create, pass name and colors. To change a palette, pass its id and only the fields that change.

Permission: assets:write. Credits: free.

Returns: the palette with its id. The tool cannot delete a palette. Delete one in the app, with the CLI or through the API.

list_moodboards

Lists your moodboards, newest first: the name, the number of images, the description and the id. With id, it returns one moodboard with each image's link and caption. Pass an id to generate_image as moodboard_id.

Permission: assets:read. Credits: free.

Returns: a page of moodboards, or one moodboard, as GET /v1/moodboards and GET /v1/moodboards/:id return them.

save_moodboard

Creates a moodboard, or changes one. To create, pass name and images. To change a moodboard, pass its id and only the fields that change.

Permission: assets:write. Credits: free.

Returns: the moodboard with its id. Images are kept as links, never copied. The tool cannot delete a moodboard.