ByteDance Bagel
Unleash creativity with ByteDance Bagel, a powerful unified multimodal AI for image generation, editing, and understanding.
🚀Function Overview
A unified multimodal model performing image generation, image editing based on text prompts, and image understanding with optional chain-of-thought reasoning across tasks.
Key Features
- Generates images from text descriptions
- Edits images using natural language instructions
- Answers questions about image content
- Chain-of-thought reasoning for complex tasks
- Mixture-of-Transformer-Experts architecture
- Dual visual encoders for semantic and pixel-level processing
Use Cases
- •Creating images from textual prompts
- •Modifying existing images via text instructions
- •Analyzing and answering questions about visual content
- •Creative design workflows combining generation and editing
- •Visual reasoning applications
⚙️Input Parameters
prompt
stringText prompt for generation, editing, or understanding
image
stringInput image for editing or understanding tasks
task
stringTask to perform
enable_thinking
booleanEnable chain-of-thought reasoning for better results
cfg_text_scale
numberText guidance scale for how closely to follow the prompt
cfg_img_scale
numberImage guidance scale for preserving input image details
num_inference_steps
integerNumber of denoising steps
timestep_shift
numberDistribution of denoising steps between composition and details
cfg_renorm_type
stringCFG renormalization method
cfg_renorm_min
numberMinimum CFG renorm value
seed
integerRandom seed for reproducible results
output_format
stringOutput image format
output_quality
integerImage compression quality for lossy formats
💡Usage Examples
Example 1
Input Parameters
{
"task": "image-editing",
"image": "https://replicate.delivery/pbxt/N3igSnChiyUgl9AKnXNhg5wKnyVn05z03gf4JaxtPO4EeHgs/image.png",
"prompt": "Change the cosmic cloud background of the floating temple to a clear blue sky with a gentle sunrise on the horizon. Keep all temple architecture, figures, and other elements exactly as they are.",
"cfg_text_scale": 4,
"cfg_img_scale": 1.5,
"output_format": "webp",
"cfg_renorm_min": 1,
"output_quality": 90,
"timestep_shift": 3,
"cfg_renorm_type": "global",
"enable_thinking": false,
"num_inference_steps": 50
}Output Results
Quick Actions
Technical Specifications
- Hardware Type
- L40S
- Run Count
- 3.6k
- Commercial Use
- Supported
- Platform
- Replicate