GPT-Image Pipe for Open WebUI

GPT-Image Pipe is a pipe for Open WebUI that connects you directly to OpenAI's advanced image generation models. This pipe supports both text-to-image generation and image-to-image editing within a conversational interface.

✨ Features

• Multi-Model Support: fast access to gpt-image-1 and gpt-image-1.5.
• Conversational Generation: Generate images based on your chat context.
• Image Editing: Seamlessly edit images by uploading them or referencing generated ones in the chat.
• Configurable Options: Fine-tune your experience with adjustable parameters for image count, size, quality, and more.

🚀 Installation

This component is designed to run within the Open WebUI ecosystem as a Pipe.

1. Ensure you have Open WebUI installed and running.
2. Navigate to the Functions section in your Open WebUI administration panel.
3. Create a new function.
4. Paste the contents of gpt_image_pipe.py into the code editor.
5. Restart Open WebUI for the dependencies to install.
6. Save and activate the function.

⚙️ Configuration (Valves)

You can configure the behavior of the pipe using "Valves" in the Open WebUI interface.

Valve	Description	Default
OPENAI_API_KEYS	Required. Your OpenAI API Key(s). Supports multiple comma-separated keys for load balancing.	""
IMAGE_NUM	Number of images to generate per request (1-10).	1
IMAGE_SIZE	Dimensions of the generated image (1024x1024, 1536x1024, 1024x1536, or auto).	auto
IMAGE_QUALITY	Quality setting (high, medium, low, auto).	auto
MODERATION	Moderation strictness (auto or low).	auto
INPUT_FIDELITY	(gpt-image-1 only) Effort to match source style/features (high or low).	low

🎨 Usage

Text-to-Image Generation

1. Select GPT Image 1 or GPT Image 1.5 from the model dropdown in your chat.
2. Type a description of the image you want to create.

   "A futuristic cityscape with flying cars and neon lights, cyberpunk style."

Image-to-Image Editing

1. Upload an image to the chat or reference a previously generated image.
2. Provide a prompt describing how you want to modify the image.

   "Make it look like a watercolor painting." "Add a giant robot in the background."

📜 License

This project is licensed under the MIT License.