Text To Image

Let's get the important thing out of the way: this node does not generate images from a text prompt. No model, no diffusion, no KSampler. It types your text onto a canvas with PIL, like a caption generator or a sign-maker, and hands you an IMAGE you can feed into the rest of your workflow. "Text To Image" here means literally rendering text to an image. If you googled this expecting Stable Diffusion text-to-image, that's the trap - close the tab and look for a Checkpoint Loader plus KSampler instead. If you actually need a text rendered as pixels, keep reading, because that's genuinely useful.

What it actually does

You give it text, it renders it with a truetype font onto a dynamically sized canvas, and outputs an IMAGE tensor. The canvas grows to fit the text, wrapped intelligently - it's CJK-aware, so mixed Chinese/English lines wrap at sensible points rather than mid-character. It ships a Songti (宋体) font in the pack's fonts/ folder for exactly this, and the font dropdown picks it up automatically (along with a "default" system font). Output is capped at a 1024px max width so it stays pipeline-friendly.

Inputs that matter

• text - multiline. The default is Hello World\n你好世界, which tells you everything about who built this.
• font_name - 默认字体 or Songti. Adding your own .ttf/.ttc/.otf to the pack's fonts/ folder makes it appear here.
• font_size - 12–200, default 48.
• text_color / background_color - hex strings like #000000 / #FFFFFF. Both 3- and 6-digit hex work.
• padding - margin in pixels around the text, 0–100, default 30.
• line_spacing - line-height multiplier, 0.5–3.0, default 1.2.

Outputs

One output: image (IMAGE tensor). It's OUTPUT_NODE = True, so you'll see it in the UI as an output/preview. From there it can go anywhere an image goes: VAE Encode for img2img, a Combine Image+Audio node as a frame, or a text-overlay into a collage workflow. Making a title card for an AnimateDiff clip or watermarking a batch is exactly the sweet spot.

Gotchas

The word-wrap logic does the heavy lifting but maxes out - very long unbroken strings get wrapped by character, which for a 1024px cap is fine until it isn't. Colors are parsed strictly: garbage hex silently defaults to black, which can look like the node "did nothing." And there's no transparency option - background is always a solid color, so you can't get a text-only overlay with alpha from this. Use a compositing node to punch the text over an image instead.

Installing

Part of the ZMG pack. ComfyUI Manager → search "ZMG" / "ComfyUI-ZMG-Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/vanche1212/ComfyUI-ZMG-Nodes
cd ComfyUI-ZMG-Nodes
pip install -r requirements.txt

Restart, find it under ZMGNodes/image. Needs Pillow (for the rendering) and torch (to build the tensor), both in the pack's requirements. Useful tool - just remember the name is doing something different from what you think.