Text-to-Image Processor

Overview

The Text-to-Image Processor (text_to_image) generates a brand-new PNG image from a text prompt. It supports:

• Multiple model backends, including:

  • SDXL (sdxl)
  • Flux Schnell (flux)
  • Flux v1.1 Pro (flux2)
  • Recraft v3 (recraft)
  • Imagen (imagen, imagen-fast, imagen-ultra)

• Positive (prompt) and negative (negative_prompt) guidance

• Fixed sizing (width and height in pixels)

• Free-form overrides via options_json—all values are sanitized and clamped before calling the model

---

Example Usage

- processor:
    processor: "text_to_image"
    params:
      prompt: "A samurai standing under cherry blossoms"
      negative_prompt: "blurry"
      model: "sdxl"
      seed: 42

      width: 800
      height: 600

      options_json: |
        {
          "num_inference_steps": 28,
          "guidance_scale": 3.5,
          "aspect_ratio": "4:3"
        }
    outputs:
      target: "mem/samurai.png"

---

Inputs

This processor does not require a source image.

Key	Required	Description
—	—	N/A

---

Params

Key	Type	Required	Default	Description
prompt	string	Yes	—	Positive text describing the desired image.
negative_prompt	string	No	""	Text to steer the model away from unwanted content.
model	string	No	"sdxl"	Model code (e.g. "flux", "flux2", "sdxl", "recraft", "imagen", "imagen-fast", "imagen-ultra").
seed	int	No	(rand)	Random seed for reproducibility.
width	int	Yes	—	Output width in pixels.
height	int	Yes	—	Output height in pixels.
options_json	string	No	"{}"	JSON string of additional model-specific inputs; allowed keys vary by model and are safely filtered.

---

Outputs

Key	Required	Description
target	Yes	Path where the PNG is written.

---