Flux family
# Qwen Inpaint Advanced AI-powered image editing and selective inpainting. ## Overview Qwen Inpaint is a sophisticated multimodal image editing tool designed to perform precise, context-aware modifications to existing visual assets. By leveraging the advanced reasoning capabilities of the Qwen architecture, this model allows users to mask specific areas of an image and replace or modify them based on natural language prompts. Unlike traditional generative models that require full-image regeneration, Qwen Inpaint maintains the structural integrity and stylistic consistency of the original image while seamlessly blending new elements into the masked region. This model is particularly distinct due to its ability to interpret complex visual instructions. Whether you are looking to swap an object, change a background element, or modify specific textures within a composition, Qwen Inpaint analyzes the surrounding pixels to ensure that lighting, shadows, and perspective are preserved. It serves as an essential utility for marketing teams and creative agencies that need to iterate on high-quality assets without starting from scratch. ## Use Cases * **Product Photography Refinement:** Swap out product colors or textures while keeping the original studio lighting and background intact. * **Dynamic Ad Variations:** Quickly generate multiple versions of a campaign image by swapping out props or seasonal elements within a scene. * **Asset Localization:** Replace text or culturally specific imagery within an existing ad creative to better suit different global markets. * **Background Cleanup:** Remove unwanted objects, distractions, or people from high-quality stock photography to create a cleaner, more focused composition. * **Creative Prototyping:** Experiment with different visual concepts by modifying specific segments of a hero image during the brainstorming phase. * **E-commerce Optimization:** Update lifestyle images to include new seasonal accessories or apparel without the need for additional photoshoots. ## Parameters This model offers a robust set of parameters to control the generation process. Users can adjust the 'strength' to determine how much the model deviates from the original pixels, set the 'guidance_scale' to balance prompt adherence, and choose from various 'image_size' presets to match specific ad platform requirements. Please refer to the input schema for the full list of configuration options including seed control, acceleration modes, and safety settings. ## Tips for Best Results * **Mask Precision:** Ensure your mask covers the target area completely, but avoid over-masking. A tighter mask around the object of interest often yields more natural blending. * **Descriptive Prompts:** Use specific, descriptive language in your prompt. Instead of "change the object," use "a sleek, modern silver watch on the wrist." * **Manage Strength:** Use a lower strength value (e.g., 0.7-0.8) for subtle edits and a higher strength value (e.g., 0.9-1.0) when you want the model to generate more significant changes. * **Negative Prompting:** Utilize the negative prompt field to explicitly exclude unwanted elements, such as "blurry," "distorted," or "low quality," to refine the output clarity. * **Iterative Testing:** If the first result isn't perfect, adjust the seed or the guidance scale slightly to explore variations while maintaining the core composition. ## About Qwen Inpaint is powered by the fal-ai/qwen-image-edit/inpaint endpoint, utilizing the cutting-edge Qwen multimodal architecture. Developed to bridge the gap between complex reasoning and visual generation, this model represents the latest in generative AI image editing technology. It is designed for professional creative workflows, offering high-performance inference capabilities suitable for commercial marketing and advertising production.
Model input reference derived from preset schema.
| Name | Type | Required | Default | Description |
|---|---|---|---|---|
| seed | number | No | null | The same seed and the same prompt given to the same version of the model will output the same image every time. |
| prompt | string | Yes | "" | The prompt to generate the image with |
| mask_url | image | Yes | "" | The URL of the mask for inpainting |
| strength | number | No | 0.93 | Strength of noising process for inpainting |
| image_url | image | Yes | "" | The URL of the image to edit. |
| sync_mode | boolean | No | false | If `True`, the media will be returned as a data URI and the output data won't be available in the request history. |
| image_size | select | No | "landscape_4_3" | The size of the generated image. |
| num_images | number | No | 1 | The number of images to generate. |
| acceleration | select | No | "regular" | Acceleration level for image generation. Options: 'none', 'regular'. Higher acceleration increases speed. 'regular' balances speed and quality. |
| output_format | select | No | "png" | The format of the generated image. |
| guidance_scale | number | No | 4 | The CFG (Classifier Free Guidance) scale is a measure of how close you want the model to stick to your prompt when looking for a related image to show you. |
| negative_prompt | string | No | " " | The negative prompt for the generation |
| num_inference_steps | number | No | 30 | The number of inference steps to perform. |
| enable_safety_checker | boolean | No | true | If set to true, the safety checker will be enabled. |