Seedream 5.5: ByteDance Promises 100 Successive Edits Without Drift and Native Transparent Backgrounds
ByteDance Seed has launched Seedream 5.5, its new image model featuring serial editing that preserves untouched areas down to the pixel, native RGBA output with transparent backgrounds, and sharper text. Dreamina has rolled it out to all users, and a Pro version has been announced.
Editing a generated image one hundred times in a row without degrading its quality: that is the promise behind Seedream 5.5, the new image model from ByteDance, unveiled this Monday, November 9, by its Seed research team. The model also introduces native transparent background output and sharper text rendering. Dreamina, the group’s creative app (CapCut), promptly opened access to all its users.
Three New Features Built for Production
Serial editing. When performing consecutive edits, each image serves as the input for the next, causing small errors to compound: colors drift, fine details fade, and faces change. ByteDance claims to have trained Seedream 5.5 to preserve pixel-perfect fidelity in unedited areas while eliminating color shifts. Its demonstration shows 100 consecutive edits on a single portrait (clothing, jewelry, tabletop items), followed by floral arrangements and a room redecorated step by step, all while keeping perspective, shadows, and occlusions consistent.
Native transparent backgrounds. The model outputs direct RGBA images, complete with an alpha channel offering full or partial transparency. In practice, an object, character, or decorative title comes out ready to drop onto a mockup, with no cutout or edge-cleaning required. ByteDance highlights owl feathers, hair, and sheer fabrics—areas where traditional cutout tools stumble most often.
Realism and fine print. Lighting, materials, and skin textures are described as more natural, reducing “oily skin” artifacts and overexposure. Text rendering also takes a step forward, including complex Chinese characters rendered in small font sizes.
Big Numbers, but All In-House
The published benchmarks come straight from ByteDance. On its MagicArena evaluation platform, evaluators scored the model across six use cases: e-commerce, marketing, film and television, gaming, office productivity, and fashion. Seedream 5.5 improved across the board compared to Seedream 5.0 Pro, with the proportion of images deemed usable more than doubling in film/television and office productivity. ByteDance reports a severe text corruption rate of 0.4% (1.7% for mild flaws) and claims an edge over GPT Image 2.5 Sunburst and Nano Banana 2.1 in text structure and text-to-image realism.
These comparisons rely on preferences from experts selected by the developer. No independent benchmarks have been published yet. The model was not yet listed on the Arena text-to-image leaderboard—which is based on user blind votes—when checked on November 9 at 1:00 PM (Paris time): the top five spots were held by versions of OpenAI’s GPT-Image, led by GPT-Image-2.5 Sunburst (1,428 points), while the earlier Seedream 5.0 Pro stood in 17th place (1,249 points). These rankings fluctuate with every update.
Availability and Pricing
ByteDance Seed announced the model for Dreamina, CapCut, and BytePlus, its international cloud platform. Dreamina already offers it to all users under a limited-time promotion: unlimited usage for 7 days for everyone and 30 days for Basic subscribers and above. In China, the model is rolling out gradually on Jimeng, Doubao, and Volcano Ark, according to IT之家. The same outlet reports that the API bills input and output images separately, with no charge for text tokens, and that editing a 2K square image costs around 0.28 yuan—less than half the price of Seedream 5.0 Pro. ByteDance did not include this pricing in its official announcement. A Pro version in the 5.5 series has been promised, though no release date was given.
Our take
ByteDance is not chasing the prettiest single-shot image: it is tackling what actually drains designers’ budgets—the endless back-and-forth iterations. Leaving untouched areas intact and delivering pre-cutout assets are precisely what make an image model viable in a real production pipeline (product catalogs, ad variations, mockups). What remains to be seen, beyond curated demos, is whether stability holds up across 100 rounds with everyday prompts, and whether the comparisons against GPT Image and Nano Banana stand up to independent testing. If the pricing reported in China holds true globally, it will make Seedream a formidable contender for high-volume workflows.
Sources
- ByteDance Seed — Introducing Seedream 5.5 (November 9, 2026) Primary source
- ByteDance Seed — Seedream 5.5 page Primary source
- ByteDance Seed on X — Introducing Seedream 5.5 Primary source
- Dreamina on X — Seedream 5.5 makes its global debut Primary source
- Arena — Text-to-Image Leaderboard (accessed November 9, 2026) Secondary source
- IT之家 — 字节跳动发布新一代图像创作模型 Seedream 5.5 (November 9, 2026) Secondary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
