Step Image Edit 2 Image Model Released, Leading Performance and Efficiency
TL;DR · AI Summary
Step Image Edit 2 ranks #1 on KRIS-Bench with 3.5B parameters, outperforming models 5-6x larger, at 0.7s per image generation.
Key Takeaways
- 3.5B-parameter model ranks #1 on KRIS-Bench, outperforming 5-6x larger models.
- Text-to-image generation: 0.7s/image, editing: 1.6s/edit, cost: $0.003/image.
- Supports accurate bilingual text rendering (Chinese + English) and style transfe
Outline
Jump quickly between sections.
Step Image Edit 2 model launched with 3.5B parameters, ranking #1 on KRIS-Bench.
KRIS-Bench tests show top ranking across overall, factual, and conceptual categories.
Text-to-image generation takes 0.7s/image, editing 1.6s/edit, cost $0.003/image.
Supports text-to-image generation, instruction-based editing, bilingual rendering, and style transfer.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- Step Image Edit 2模型
- 性能优势
- KRIS-Bench #1排名
- 5-6倍参数效率
- 功能特性
- 文本生成图像
- 指令编辑
- 中英双语渲染
Highlights
Key sentences worth saving and sharing.
Step Image Edit 2 ranks #1 on KRIS-Bench with 3.5B parameters, outperforming models 5-6x larger.
Inference speed: 0.7s for text-to-image, 1.6s for editing, cost $0.003 per image.
Supports accurate bilingual text rendering (Chinese + English) and subject-consistent style transfer.
A 3.5B-parameter image model that ranks #1 on KRIS-Bench — the leading benchmark for instruction-based image editing — across overall, factual, and conceptual categories. It outperforms models at 5–6x the parameter count.
Performance
- 0.7s https://t.co/gsSdrj25IJ
Step Image Edit 2 is live!! A 3.5B-parameter image model that ranks #1 on KRIS-Bench — the leading benchmark for instruction-based image editing — across overall, factual, and conceptual categories. It outperforms models at 5–6x the parameter count. Performance - 0.7s text-to-image - 1.6s per edit - $0.003 per image Capabilities → Text-to-image generation → Instruction-based image editing → Bilingual text rendering (accurate Chinese + English characters) → Style transfer with subject consistency Get started Model ID: step-image-edit-2 Available now on the Stepfun open platform. Docs→platform.stepfun.ai/docs/en/guides
