Seedream V4.5

• 🆕 ByteDance's editing variant of the Seedream 4.5 image family.
• 🔧 Text-guided editing through natural-language instructions, no manual masking.
• 🎯 More accurately locks main subjects across multi-image edits.
• 📚 Enhanced dense text and typography rendering.
• 🧠 All-round improvement over Seedream 4.0 via model scaling.
• 🏢 Built by the ByteDance Seed team.
• ⚡ Unified architecture merging generation and editing.

ByteDance

ByteDance is a Chinese internet technology company headquartered in Beijing, widely known as the parent company behind TikTok and Douyin. Beyond social media, ByteDance has invested heavily in artificial intelligence research, building generative media models…

Seedream V4.5 (edit) is the image-editing side of ByteDance's Seedream 4.5 family, sharing architecture and training foundations with the text-to-image model so that generated and edited assets stay visually consistent within a project. Built by the ByteDance Seed team, it merges text-to-image synthesis and precise editing into a single unified architecture, handling edits through natural-language instructions rather than manual masking.

According to ByteDance's own description, Seedream 4.5 achieves an all-round improvement through overall model scaling: it more accurately identifies and locks main subjects in multi-image edits, preserves reference details and textures, and enhances dense text rendering. On the team's internal MagicBench, ByteDance reports gains over Seedream 4.0 across prompt adherence, alignment, and aesthetics.

Within Venice's catalog, this editing model sits alongside the standard image generator Seedream V4.5 and the lighter Seedream V5 Lite editing variant, while the broader Seedream lineage extends into ByteDance's Seedance 2.0 video models.