ComfyUI Ideogram Image and Text Tools

Generate a logo. Build a sticker sheet. Export every size you need. All without leaving ComfyUI.

⚠️ Before you wire anything up: every node here uses the inverse of ComfyUI's default mask convention. Read Core Concept: the Alpha Convention first, or you'll get silently wrong output, not an error.

Why This Exists

This repository turns generated images — from Ideogram or anywhere else — into production-ready creative assets: sticker sheets, logo packages, wordmarks, multi-size exports, all as composable ComfyUI nodes.

The Ideogram ecosystem already has prompt builders, JSON builders, layout tools, character tools, background removal solutions, and generic image editing tools. This repository does not duplicate any of that — it focuses on the gap after generation: turning a finished image into something you can actually ship.

Concretely: generated images get prepared (AlphaPrep, WordmarkGenerator) and then packaged (StickerSheetBuilder, LogoAssetBuilder, AssetPackExport).

Quickstart

The minimum useful chain — five minutes to a real exported asset, and the foundation every other workflow in this repo builds on:

LoadImage -> AlphaPrep: Mask Adapter -> AlphaPrep: Trim
  -> AlphaPrep: Mask Adapter -> JoinImageWithAlpha -> SaveImage

This loads any image, strips its transparent border, and saves it back out with correct alpha.

For something to actually open and look at, workflows/ has three polished, ready-to-load templates — each with a yellow README note built into the graph explaining what it does and what to change:

• Sticker Prep Pipeline — this quickstart chain, plus outline/shadow/resize.
• Sticker Sheet from Three Logos — batch multiple assets into one packed sheet.
• Full Brand Kit from One Logo — the most complete single-graph tour: LogoAssetBuilder, AssetPackExport, and ThumbnailLegibilityCheck together.

examples/ also has a flat-format workflow per node system, used by this repo's own test suite — see workflows/README.md for how the two folders differ.

Status

All six node systems below are implemented, each with an example workflow in examples/ and verified end-to-end against a live ComfyUI instance. See CHANGELOG.md for version history and docs/ for design notes and known limitations.

Node Systems

1. AlphaPrep (implemented) — trim transparent borders, expand the canvas by explicit per-edge padding, pad, center, resize canvas, generate sticker outlines and drop shadows, preview against backgrounds, and adapt the mask convention at ComfyUI core boundaries. See docs/nodes/alphaprep.md.

   

   Trim → Outline → Drop Shadow → Resize Canvas → Preview Background, chained.

   

   Canvas Expand adds asymmetric per-edge padding (here: 40px top, 60px left, 10px bottom/right) — the outline marks the original trimmed bounds.

2. StickerSheetBuilder (implemented) — pack multiple transparent assets into print-ready sticker sheets with configurable layouts, margins, and sheet sizes. See docs/nodes/stickersheetbuilder.md.

   

   Five hotdog variants packed with the packed (shelf) layout.

3. WordmarkGenerator (implemented) — typography-first branding asset generation (band logos, product names, podcast branding, etc.). This renders text directly with Pillow and a font file — no Ideogram or other image-generation call involved. On Linux, pass an explicit font_path: the bundled fallback font names default to Windows/macOS fonts and silently degrade to a low-fidelity bitmap font otherwise (see Known Limitations). See docs/nodes/wordmarkgenerator.md.

   

   Rendered directly from text with the wide style preset — no image generation involved.

4. LogoAssetBuilder (implemented) — full logo asset packages: variants, transparent exports, square/banner versions, monochrome versions. See docs/nodes/logoassetbuilder.md.

   Transparent	Square	Banner	Monochrome

   One logo asset in, four production-ready package outputs out.

5. AssetPackExport (implemented) — export one asset at multiple named target sizes (e.g. icon:128x128, square:512x512, banner:1500x500) in a single pass, one execution per requested size via ComfyUI's list-output mechanism. See docs/nodes/assetpackexport.md.

   

   One sizes spec (icon:96x96, square:200x200, banner:480x160), three exports, one execution.

6. ThumbnailLegibilityCheck (implemented) — render an asset at multiple small sizes side by side, each at real pixel dimensions, to check whether a logo/sticker survives at icon/favicon scale before shipping it. See docs/nodes/thumbnaillegibilitycheck.md.

   

   Each thumbnail is rendered at its real pixel size, not scaled up — this is what it actually looks like that small.

Design Principles

• Follow standard ComfyUI node conventions; no custom image formats.
• Every node solves a real production problem — no novelty or gimmick features.
• Small, focused, composable nodes over giant all-in-one nodes.
• Asset workflows first, API wrappers second.
• Works with any image source where possible, not just Ideogram.

Non-Goals

This repository will not implement prompt generators, prompt mutators, shot planners, JSON builders, layout/character planners, dataset tools, or LoRA/training tools. Those belong in other repositories.

Background removal is a deliberate non-priority too — it's not the primary value of this repository, and may be added later only as a supporting utility, not a lead feature.

Core Concept: the Alpha Convention

Every node in this repository treats IMAGE + MASK as a single transparent asset, where MASK is the asset's alpha channel: 1.0 = opaque content, 0.0 = fully transparent. All six node systems — AlphaPrep, StickerSheetBuilder, WordmarkGenerator, LogoAssetBuilder, AssetPackExport, and ThumbnailLegibilityCheck — depend on this shared contract; it's the one thing to understand before wiring nodes together.

This is the opposite of ComfyUI's own inpainting-mask convention (where 1.0 means "masked out"). Concretely: LoadImage's MASK output and JoinImageWithAlpha's alpha input both expect the inpainting-style convention, not this package's. Use AlphaPrep: Mask Adapter at both boundaries — LoadImage -> AlphaPrep: Mask Adapter -> any node here, and any node here -> AlphaPrep: Mask Adapter -> JoinImageWithAlpha -> SaveImage. See docs/README.md for the full explanation, and examples/ for working reference wiring.

Known Limitations

• Mask convention mismatch with core ComfyUI nodes (see above) — forgetting AlphaPrep: Mask Adapter at either boundary produces silently wrong output, not an error.
• WordmarkGenerator font fallback on Linux: if no font_path is supplied and none of the fallback system font names (DejaVuSans-Bold.ttf, DejaVuSans.ttf, arial.ttf) resolve on the host, it silently falls back to Pillow's low-fidelity built-in bitmap font. arial.ttf in particular is a Windows/macOS font name and will not resolve on most Linux ComfyUI installs. Supply an explicit font_path for predictable typography on Linux.

Installation

Clone this repository into your ComfyUI custom_nodes/ directory and restart ComfyUI:

cd ComfyUI/custom_nodes
git clone https://github.com/SurrealByDesign/ComfyUI_Ideogram_Image_and_Text_Tools.git

No extra dependencies to install — this package only needs torch, numpy, and Pillow, all of which ComfyUI already provides.

Most recently tested end-to-end against ComfyUI 0.24.0. The nodes are plain Python with no version-specific API usage, so other recent ComfyUI versions are likely fine, but 0.24.0 is the only version this has actually been run against — this line will be updated as newer versions get verified.

Testing

pytest

Contributing

See CONTRIBUTING.md for coding and testing standards.

License

See LICENSE.