Ideogram v4: Text to Image (API)
Input a text prompt in structured JSON format. Generate a high-quality image with precise layout control, text rendering, and color palette support.
试试一个示例
关于此工作流
This ComfyUI tutorial workflow demonstrates how to drive Ideogram v4 via a structured JSON prompt to get clean text rendering, consistent brand colors, and reliable layout control. Instead of a single free-form sentence, you build a compact JSON object that describes the scene, in-image text, a color palette, and per-object bounding boxes. That JSON is fed into the IdeogramV4 partner API node, which returns a high-quality image that adheres to your specified layout as closely as the model allows. The output flows directly to SaveImage.
To make authoring easier, the JSON Prompt Builder group (enable it by selecting the group and pressing Ctrl+B) includes a MarkdownNote with a schema/template and an optional GeminiNode that can transform a plain brief into a well-formed JSON prompt. You can preview the structured prompt with PreviewAny before sending it to IdeogramV4, iterate on bounding boxes or palette values, and then render the final image. This approach is ideal when you need reliable in-image text, brand-safe palettes, and object placement that a natural-language prompt alone struggles to achieve.
常见问答
You can enter plain natural language directly into the IdeogramV4 node. However, this workflow is optimized for a structured JSON prompt, which gives much stronger control over in-image text, color palette, and object placement via bounding boxes.
Use the template in the MarkdownNote as your source of truth. A minimal example might include: prompt (overall description), palette (array of hex colors), and layout items with type, text/subject, and a bbox [x, y, w, h] in normalized 0–1 coordinates. Example (keys may vary by template): { "prompt": "retro poster of a mountain sunrise", "palette": ["#0F0F0F", "#FF6A00", "#FFD54D"], "layout": { "items": [ {"type": "text", "text": "SUMMER FEST", "bbox": [0.1, 0.05, 0.8, 0.2]}, {"type": "image", "subject": "mountain sunrise", "bbox": [0.1, 0.28, 0.8, 0.55]} ] } }
Increase padding in the text item’s bounding box, reduce the text length, and avoid overlapping boxes. Keep bbox values within 0–1, ensure items don’t collide, and iterate—small bbox adjustments often resolve clipping and alignment issues.
Define a concise palette array of your brand hex colors in the JSON. Limit it to a few key colors to nudge the model strongly. If results drift, simplify the palette, reduce conflicting style cues, and re-run. The model aims to respect your palette while balancing overall image quality.