How to Use LLaDA-Image Online

This tutorial shows how to use LLaDA-Image through the independent browser workspace on this site. It is not an official inclusionAI manual. For sampling hyperparameters, follow the public repository.

Disclaimer: We provide online UI access and credit-based generation. Official code and weights remain on GitHub, Hugging Face, and ModelScope.

Before you start

• Decide whether you need generation (text only) or editing (reference image).
• Skim How to Use LLaDA-Image for the short product overview.
• Keep official GitHub open for Diffusers tips.

Step 1 — Open the workspace

Go to LLaDA-Image Online and click into the workspace. Confirm the model label: LLaDA-Image (Base-oriented) vs LLaDA-Image Turbo.

Step 2 — Generate from text

1. Write a specific prompt: subject, setting, lighting, aspect ratio.
2. If you need poster copy, quote exact Chinese/English strings.
3. Generate once, then change one variable at a time.

Example starter:

Photoreal ceramic mug on a sunlit oak table, soft window light,
shallow depth of field, no watermark, product photography

More templates: LLaDA-Image prompts.

Step 3 — Edit with an instruction

1. Upload a reference image.
2. State what must stay (identity, logo, pose).
3. State the single change (background, color, text).

Example:

Keep the same person and clothing. Change only the background
to a rainy Tokyo street at night. Preserve facial identity.

Dedicated page: LLaDA-Image Editor.

Step 4 — Choose Turbo when iterating

Use LLaDA-Image Turbo for faster drafts. Switch back to Base when you need richer detail. Full write-up: Turbo vs Base.

Step 5 — Credits and limits

Generation uses site credits. Compare plans on Pricing. Do not assume unlimited free inference.

Official citations

• GitHub inclusionAI/LLaDA-Image
• Hugging Face checkpoints
• arXiv:2609.03796

CTA

Open the workspace and run one generate + one edit pass today.