Seedream 5.5 Uncensored Advanced

Overview

Seedream 5.5 Uncensored Advanced is an image generation and editing model by ByteDance. It turns a written prompt into a new image, or applies the prompt as an edit to your reference images. It’s built for repeatable edits, so you can refine one image over many rounds with less drift. It can also produce RGBA outputs, which helps when you need clean cutouts without manual background removal.

What you can build

• Product cutouts and catalog images with transparent backgrounds
• Character-consistent concept art iterations across multiple edit rounds
• Poster and banner concepts with clearer small text than older image models
• Fashion, editorial, and lifestyle photo-style generation from text
• Multi-reference compositing, like combining pose, outfit, and background references
• Sticker packs, icons, and UI illustration assets with alpha transparency

Inputs

• A required written prompt that describes what to generate or how to change your references. ByteDance docs recommend staying under about 600 English words or 300 Chinese characters.
• Optional reference images you want to edit or combine. You can provide up to 10 images per run. Use common formats like JPEG, PNG, or WEBP.
• A required output width in pixels. The allowed width depends on total pixel area and aspect ratio limits.
• A required output height in pixels. The width × height area must be between 921,600 and 4,624,220 pixels.
• Optional background behavior. You can let the model decide, force an opaque background, or request a transparent background.
• Optional output file format selection. You can request JPEG or PNG.
• Optional watermark toggle. When enabled, the output includes an “AI generated” mark in the lower-right corner.

Outputs

• One generated image in JPEG or PNG format.
• If you request a transparent background, the output is a PNG with an alpha channel (RGBA).
• The response includes a downloadable image file and basic file metadata, such as MIME type and file size.

Recommended settings

• For stickers, icons, and product cutouts: request a transparent background and choose PNG.
• For long multi-round edit chains: keep the background opaque while you iterate, then switch to transparency at the end if needed.
• For extreme aspect ratios: stay within a 1:16 to 16:1 width-to-height ratio to avoid rejected requests.

Limitations

• You can provide up to 10 reference images per run. Extra images may be ignored on some deployments.
• Output dimensions must follow both pixel-area limits (921,600 to 4,624,220 pixels) and aspect ratio limits (1:16 to 16:1).
• Transparent output requires PNG. JPEG can’t store transparency.
• Low-quality inputs reduce edit accuracy. Blurry photos, heavy compression, screenshots, or noisy scans can cause artifacts.
• If you want stable multi-turn editing, you must feed the last output image back as the next reference.

Safety & compliance

• Some hosts apply safety checks to prompts and reference images. A request may be blocked even if the model can generate it.
• Don’t upload images you don’t own or have rights to use.
• Don’t generate illegal sexual content. This includes any sexual content involving minors.
• Don’t use the model to create non-consensual intimate imagery or to impersonate real, identifiable people.