Simply make AI models faster, cheaper, smaller, greener!

---

We can optimize any diffusers models and optimized FLUX.1-dev using the following techniques:

---

Run an optimized FLUX.1-dev as a serverless endpoint to generate images.

⚠️ Important Notes:

• Compilation Time: The first request may take 2-3 minutes as the model compiles for optimal performance
• Warmup Time: Subsequent requests will be faster but may still have a brief warmup period

---

---

Usage

The worker accepts the following input parameters:

Parameter	Type	Default	Required	Description
prompt	str	None	Yes	The main text prompt describing the desired image.
negative_prompt	str	None	No	Text prompt specifying concepts to exclude from the image
height	int	1024	No	The height of the generated image in pixels
width	int	1024	No	The width of the generated image in pixels
seed	int	None	No	Random seed for reproducibility. If None, a random seed is generated
num_inference_steps	int	25	No	Number of denoising steps for the base model
guidance_scale	float	7.5	No	Classifier-Free Guidance scale. Higher values lead to images closer to the prompt, lower values more creative
num_images	int	1	No	Number of images to generate per prompt (Constraint: must be 1 or 2)

Example Request

{
  "input": {
    "prompt": "a knitted purple prune",
    "negative_prompt": None,
    "height": 1024,
    "width": 1024,
    "num_inference_steps": 25,
    "guidance_scale": 7.5,
    "seed": 42,
    "num_images": 1
  }
}

which is producing an output like this:

{
  "delayTime": 11449,
  "executionTime": 6120,
  "id": "447f10b8-c745-4c3b-8fad-b1d4ebb7a65b-e1",
  "output": {
    "image_url": "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAABAAAAAQACAIAAADwf7zU...",
    "images": [
      "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAABAAAAAQACAIAAADwf7zU..."
    ],
    "seed": 42
  },
  "status": "COMPLETED",
  "workerId": "462u6mrq9s28h6"
}

and when you convert the base64-encoded image into an actual image, it looks like this: