Upload a reference image, describe the change you want, and run the dedicated LongCat-Image-Edit model in your browser.
Use it for object changes, Chinese or English text replacement, style shifts, lighting adjustments, and controlled visual revisions.
Real image-to-image editing with the fal LongCat-Image-Edit model
No Images Generated
This page opens directly in image-to-image mode and selects the dedicated LongCat edit model. It is designed for controlled revisions of an existing image, not background removal, one-click upscaling, or multi-reference fusion.
Tell the model to add, remove, replace, resize, recolor, or reposition a visible object. A precise instruction that names the object, desired change, and details that must stay untouched usually produces a more controlled result than a vague request.
Replace signs, labels, packaging copy, menu text, or poster headlines. Put the exact target text in quotation marks and specify its location, color, and typography. Always inspect every character before publishing a text-heavy asset.
Shift an image toward product photography, editorial illustration, cinematic lighting, watercolor, or another visual direction while preserving the main subject. State which identity, logo, pose, or composition details must remain stable.
Adjust clothing, accessories, backgrounds, materials, colors, or presentation without rebuilding the entire scene from scratch. For identity-sensitive work, compare the face, proportions, logo, and small product details at full resolution.
The current Longcat Image integration sends one reference image to fal's LongCat-Image-Edit endpoint. If your task needs several source images combined into one scene, use a model with explicit multi-reference support instead of assuming extra uploads will be fused.
Signed-in generations are associated with your account so you can review results and download the version that works. Treat AI edits as drafts: check text, identity, trademarks, hands, edges, and untouched regions before using an output commercially.
A controlled edit starts with a clear source image and a narrow instruction. Change one major idea at a time, verify the result, then continue with a new edit if needed.
Choose a clear JPG, PNG, or supported image where the subject and target region are easy to identify. The editor intentionally accepts one reference image because the current LongCat fal path does not provide a documented multi-reference fusion workflow.
Describe what should change and what must remain. For example: Replace the white mug with a matte black mug, keep the person's hands, camera angle, background, and lighting unchanged. Put any exact in-image text inside quotation marks.
Run the edit, then inspect subject identity, text accuracy, object boundaries, reflections, shadows, and areas that were supposed to remain untouched. If the model changed too much, narrow the instruction rather than adding several new requests.
Use the best result as the next reference only when another revision is necessary. Multiple edits can accumulate visual drift, so preserve an original copy and compare each round against it before downloading the final image.
Practical limits, pricing behavior, prompts, and model selection.
Upload one reference image, make one precise change, and inspect the result before continuing.