# Skill: Render and edit readable text in images with Qwen-Image

> Produce images that contain clean, legible text (posters, covers, signage, UI mockups) and edit text/objects in existing images, using the open-weights Qwen-Image / Qwen-Image-Edit models in ComfyUI.

Source: sodigi-learn - comfyui/qwen-image - https://sodigi.io/learn/comfyui/qwen-image

## When to use

- The image must contain readable words (headline, label, sign, button text) - where most other open models produce gibberish.
- You need to change text or an object in an already-generated image without regenerating everything.
- A user wants a local, free alternative to the closed Ideogram API.
- Building posters, channel/podcast covers, packaging, memes with captions, or app UI mockups.

## Core rules

- Use Qwen-Image (Alibaba, 20B, Apache-2.0) when legible text matters; it handles English, Russian and Chinese text.
- For images with NO text, FLUX.2 is usually nicer/faster - Qwen-Image's edge is specifically text.
- Use Qwen-Image-Edit to modify an existing image: replace text, swap/remove objects, compose from 1-3 inputs, non-destructively.
- Put the exact desired text in quotes in the prompt so the model renders those letters, not invented ones.
- Qwen-Image is natively supported in ComfyUI (use the built-in templates); low VRAM -> use a GGUF/quantized build.
- It renders headlines and short strings well, not full pages of body text.
- Ideogram is a closed API and cannot run in ComfyUI; Qwen-Image is the open local answer.

## Procedure

1. Update ComfyUI to a recent version (older builds lack Qwen support).
2. Download Qwen-Image (or a GGUF build for weak GPUs) and place it as a checkpoint.
3. Open the built-in Qwen-Image template.
4. Describe the scene and specify the literal text in quotes, e.g. poster text: "Coffee on the Corner".
5. To fix/change text on an existing image, switch to Qwen-Image-Edit and edit that region only.

## Ready-to-use prompt

```
Generate this image with Qwen-Image. Render the on-image text EXACTLY as quoted, do not invent or alter wording.
Scene: <describe>
On-image text (render verbatim): "<exact text>"
Layout/style: <position, font feel, colors>
If I later need to change the text, use Qwen-Image-Edit to edit only that text region, keeping the rest of the image intact.
```

## Pitfalls

- Using Qwen-Image for text-free pictures (FLUX.2 is often better there).
- Not quoting the desired text, so the model makes up its own wording.
- Regenerating the whole image to fix one typo instead of using Qwen-Image-Edit.
- Loading the full model on a weak GPU instead of a GGUF build.
- Trying to render a full page of small body text - no model does that; use a real layout tool.
- Looking for "Ideogram in ComfyUI" - it is a closed API; use Qwen-Image instead.