Kiki Image Generation

Unleash your creativity with Kiki, a versatile image generation and editing model. Ready to experience the power of AI? Start your journey here!

Platform: Replicate

Image GenerationImage EditingLoRA IntegrationImage-to-Image Translation

91 runs

H100

License Check Required

🚀Function Overview

Versatile image generation model supporting text-to-image conversion, image-to-image transformations, and inpainting operations with customizable parameters like aspect ratio, LoRA weights, and generation quality.

Key Features

Generates images from text prompts
Performs image editing through inpainting and img2img modes
Supports aspect ratio customization and multiple resolution options
Integrates LoRA weights for specialized style/object activation
Offers multiple models (dev/schnell) with speed vs quality tradeoffs
Provides fine-grained control via prompt strength, guidance scale, and inference steps

Use Cases

•Creating original artwork from descriptive text
•Editing existing images by modifying specific regions (inpainting)
•Transforming input photos with text-guided style changes
•Generating images with customized dimensions and styles
•Exploring LoRA-driven visual concept activations

⚙️Input Parameters

prompt

string

Prompt for generated image. If you include the `trigger_word` used in the training process you are more likely to activate the trained object, style, or concept in the resulting image.

image

string

Input image for image to image or inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.

mask

string

Image mask for image inpainting mode. If provided, aspect_ratio, width, and height inputs are ignored.

aspect_ratio

string

Aspect ratio for the generated image. If custom is selected, uses height and width below & will run in bf16 mode

height

integer

Height of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation

width

integer

Width of generated image. Only works if `aspect_ratio` is set to custom. Will be rounded to nearest multiple of 16. Incompatible with fast generation

prompt_strength

number

Prompt strength when using img2img. 1.0 corresponds to full destruction of information in image

model

string

Which model to run inference with. The dev model performs best with around 28 inference steps but the schnell model only needs 4 steps.

num_outputs

integer

Number of outputs to generate

num_inference_steps

integer

Number of denoising steps. More steps can give more detailed images, but take longer.

guidance_scale

number

Guidance scale for the diffusion process. Lower values can give more realistic images. Good values to try are 2, 2.5, 3 and 3.5

seed

integer

Random seed. Set for reproducible generation

output_format

string

Format of the output images

output_quality

integer

Quality when saving the output images, from 0 to 100. 100 is best quality, 0 is lowest quality. Not relevant for .png outputs

disable_safety_checker

boolean

Disable safety checker for generated images.

go_fast

boolean

Run faster predictions with model optimized for speed (currently fp8 quantized); disable to run in original bf16

megapixels

string

Approximate number of megapixels for generated image

lora_scale

number

Determines how strongly the main LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.

extra_lora

string

Load LoRA weights. Supports Replicate models in the format <owner>/<username> or <owner>/<username>/<version>, HuggingFace URLs in the format huggingface.co/<owner>/<model-name>, CivitAI URLs in the format civitai.com/models/<id>[/<model-name>], or arbitrary .safetensors URLs from the Internet. For example, 'fofr/flux-pixar-cars'

extra_lora_scale

number

Determines how strongly the extra LoRA should be applied. Sane results between 0 and 1 for base inference. For go_fast we apply a 1.5x multiplier to this value; we've generally seen good performance when scaling the base value by that amount. You may still need to experiment to find the best value for your particular lora.

💡Usage Examples

Example 1

Input Parameters

{
  "model": "dev",
  "prompt": "A romantic scene of an Asian couple sitting on a white couch in a cozy, modern living room. The man is wearing a light denim jacket over a white t-shirt, sitting behind and wrapping his arms gently around the woman’s waist. He rests his cheek near her temple, eyes closed in a peaceful moment. The woman, named kiki, has honey-toned skin, round face, long eye lashes, double clear eye lid. She wears a white short-sleeved t-shirt and high-waisted denim shorts, leaning slightly into him and smiling warmly while looking at the camera. The background features soft lighting, wooden shelves, and evoking a romantic, cinematic atmosphere like a scene from a love story film.",
  "go_fast": false,
  "lora_scale": 1,
  "megapixels": "1",
  "num_outputs": 1,
  "aspect_ratio": "16:9",
  "output_format": "jpg",
  "guidance_scale": 2.53,
  "output_quality": 100,
  "prompt_strength": 0.79,
  "extra_lora_scale": 1,
  "num_inference_steps": 28
}

Output Results

https://replicate.delivery/xezq/LRBlozu7kurGEJCQ105IER1PkZePuh9YrZeyXNV6H2e0lRMpA/out-0.jpg

Quick Actions

Use NowView Documentation

Technical Specifications

Hardware Type: H100
Run Count: 91
Commercial Use: Unknown/Restricted
Platform: Replicate

Related Keywords

Image EditingLoRA IntegrationText-to-ImageImage-to-ImageInpaintingCustomizable Aspect RatioPrompt Strength Control

Related Models

KK777 Image Generation and Editing Model

A model for image generation and editing based on prompts and image inputs.

LamuriV3Suit

A model for generating and editing images with support for text prompts, image inputs, inpainting, and LoRA integration.

pr1modelface

A model for generating and editing images using text prompts and input images, supporting inpainting and style transfer via LoRA integration.