Espaço independente LLaDA-Image
LLaDA-Image
Novo gerador e editor de imagens open source unificado da inclusionAI. Cenas fotorrealistas, texto chinês–inglês ou edição com referência — online.
Showcase media sourced from the public LLaDA-Image repository. This site is independent and is not affiliated with inclusionAI.
Resultado de exemploModel snapshot
What LLaDA-Image ships
Figures below summarize the public Sep 2026 release notes for the 6B LLaDA-Image family. Live rankings can change.
6B
Parameters
Competitive 6B open unified generation and editing family.
50
Base steps
Recommended sampling for high-fidelity LLaDA-Image Base.
4
Turbo steps
Twin-DMD distilled LLaDA-Image Turbo for 2–4 step speed.
53.5
EN bench
Reported overall Qwen-Image-Bench score (EN 53.53 / ZH 53.38).
Why it matters
One checkpoint for generation and editing
LLaDA-Image is built as a unified diffusion stack — not a separate editing backbone bolted on later.
Unified gen + edit
A single checkpoint covers text-to-image and reference-preserving, instruction-guided editing.
Chinese–English text
Practical poster and layout rendering across bilingual creative briefs.
Instruction editing
Upload a reference and describe the change — keep subject fidelity while updating style or content.
Turbo distilled path
LLaDA-Image Turbo targets fast generation and editing in only a few sampling steps.
Generation modes
Text, VQ, and editing in one API shape
The open Diffusers pipeline supports three practical modes. This workspace mirrors the same intents online.
text
Text-to-image
Prompt-only generation with height/width divisible by 16 for Base and Turbo.
vq
VQ-conditioned
Uses LLaDA2 image VQ tokens from the prompt before diffusion — no input image required.
editing
Image editing
Provide a reference image and an instruction. Editing paths expect dimensions divisible by 32.
Evidence
Photoreal, posters, and edits
Official showcase strips highlight the three capabilities users search for after the open release.
Primary sources
Where to verify LLaDA-Image
Prefer the official open-source channels for weights, code, and the paper. This site only provides an independent browser workspace.
FAQ
LLaDA-Image questions
What is LLaDA-Image?
LLaDA-Image is a 6B open-source unified image generation and editing model family from inclusionAI, released with Base (≈50 steps) and Turbo (≈4 steps) checkpoints.
lladaimage.com é o site oficial do LLaDA-Image?
Não. É um espaço online independente para quem busca LLaDA-Image / llada image. Código e pesos oficiais estão no GitHub, Hugging Face e ModelScope.
What is LLaDA-Image Turbo?
Turbo is the Twin-DMD distilled variant for fast generation and editing in a few sampling steps. Open the Turbo page when you want the speed-oriented path.
Can I edit images as well as generate?
Yes. Upload a reference in the workspace and describe the change. The open model supports instruction-guided editing without a separate editing backbone.
Where do the showcase images come from?
The hero and gallery samples are sourced from the public LLaDA-Image repository assets, credited on-page. Live generations in the tool use this site’s hosted pipeline.
Experimente LLaDA-Image no navegador
Start from a prompt or reference image, then switch to Turbo when you need faster iterations.