Unverified60% confidenceBenchmarkExact time
The Turbo version can reduce inference steps from 50–100 steps to 4–8 steps through consistency models or progressive distillation techniques, achieving a speed improvement of more than 10x.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
9/5/2026
First Seen
Valid until: 12/4/2026
Sources
Related Entities
Related Claims
Unverified团队正在积极考虑推出 4 步或 8 步的快速 Turbo 版本，并表示模型已有内置的低步数能力，但不承诺时间点80% similarUnverifiedCompressing from 50 steps to 10 steps can theoretically reduce generation time by approximately 80%.71% similarUnverified4-step acceleration theoretically saves approximately 80% of inference time compared to 20 steps.67% similarUnverifiedThe default number of sampling steps is 12, which can be reduced to 4 or 6 to speed up generation at the cost of some image quality degradation.67% similarUnverifiedUsing GPU acceleration can speed up Ledoit-Wolf covariance shrinkage computation by 10 times.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/862249API
curl https://kongchang.com/api/v1/knowledge/claims/862249MCP
get_claim(id=862249)