Independent LLaDA-Image Turbo workspace
LLaDA-Image Turbo
The Twin-DMD distilled path of LLaDA-Image for fast generation and instruction-guided editing in about 2–4 sampling steps.
Showcase media sourced from the public LLaDA-Image repository. This site is independent and is not affiliated with inclusionAI.
결과 예시When to use Turbo
Speed without abandoning the unified stack
Use Turbo when you need rapid drafts or edit loops. Switch to Base when maximum fidelity matters more than latency.
Few-step sampling
Official guidance recommends about 4 steps with guidance scale 1.0 for Turbo inference.
Still supports editing
Turbo keeps the unified generation and editing intent — upload a reference and describe the change.
Base stays available
Need richer detail? Jump to the LLaDA-Image Base homepage for the 50-step quality path.
FAQ
LLaDA-Image Turbo questions
How is Turbo different from Base?
Turbo is a distilled checkpoint aimed at 2–4 step speed. Base targets higher fidelity around 50 steps.
Is this the official Turbo demo?
No. This page is an independent online workspace. Official weights and Diffusers code are on GitHub and Hugging Face.
Can Turbo edit images?
Yes. The open Turbo checkpoint supports generation and instruction-guided editing. Upload a reference in the workspace to try it.