Qwen Image Max Uncensored

## Overview Qwen Image Max uncensored is an instruction-following image model from Qwen Image family. You describe the result you want, and the model returns a new image that matches your instruction. With reference images, it can restyle, replace objects, adjust text in an image, or fuse multiple sources while trying to keep subject identity consistent. This helps you iterate on visuals without manual masking or pixel editing. ## What you can build - Product photo variations for catalogs, with the product kept consistent - Portrait edits that keep the same face but change wardrobe, pose, or setting - Marketing creatives that keep layout but change copy inside the image - Multi-image fusion, like “use the person from image 1 and the outfit from image 2” - Style transfer for brand shoots, like “make this look like a golden-hour editorial” - Background replacement while keeping the subject framing and lighting ## Inputs - 1 instruction written in plain language that describes what to generate or how to edit, up to 800 characters - Optional negative instruction that lists elements to avoid, up to 500 characters - Optional 1–3 reference images to edit or fuse. Refer to them as “Image 1”, “Image 2”, and “Image 3” inside your instruction. - Reference images as uploads or public URLs. Each image must be a standard raster format such as JPG, JPEG, PNG, WEBP, BMP, TIFF, or GIF. Each file must be 10 MB or smaller. - Reference image size must be within a practical range for editing. Expect best results when each side is at least 384 px. - Optional output aspect ratio choice. Auto mode matches the first reference image when present. - Optional output quality tier choice, such as a 1K or 2K tier, which affects the final pixel dimensions - Optional request for multiple output variants when you provide reference images, up to 6 images - Optional seed value (0 to 2,147,483,647) to make reruns more repeatable ## Outputs The model returns one or more generated images. - Each result is delivered as an image file in a common format like PNG, JPEG, or WEBP. - When multiple variants are requested, the response includes a list of images, one entry per variant. - The response also includes the seed used for the run, which you can reuse to reduce randomness between edits. - When the backend returns hosted URLs, they can be time-limited and may expire after about 24 hours, so download or store results promptly. ## Limitations - Reference-based editing is designed for 1–3 input images. More images can reduce instruction adherence. - Output size is constrained by the model’s maximum resolution. For Qwen Image editing models, the output is typically capped at 2048×2048, and dimensions may be adjusted to match model requirements. If you stack many unrelated changes in one instruction, the edit may drift. One main change per run is more reliable. - Low-quality inputs can break consistency. Blurry photos, heavy compression, tiny subjects, screenshots with UI chrome, or scanned images can cause artifacts or unwanted redraws. - Aspect-ratio changes can trigger reframing. If you must preserve framing, match the ratio to the reference image. ## Safety & compliance “uncensored” variants are commonly configured with weaker filtering than standard endpoints. You are responsible for lawful and compliant use. - Don’t upload images you don’t have rights to edit. - Get consent for edits of real people, especially for sensitive or sexual content. - Avoid generating illegal sexual content, sexual content involving minors, or any content that exploits or harms a person. - Don’t use this model for impersonation, non-consensual sexual imagery, or deceptive political content. - Treat reference images as sensitive data. Don’t include private information in the prompt or the image. If your workflow requires strict moderation, use a safety-checked endpoint instead of a “uncensored” configuration.