LLaDA-Image is open weights: Sume's Image API is a hosted catalog

LLaDA-Image is open weights, and Sume does not serve it. You download it from Hugging Face and run it yourself; Sume's Image API is a hosted catalog with its own model ids. If you want generation plus instruction-based editing without a GPU, openai/gpt-image-2.5 covers text-to-image and up to 16 references.

LLaDA facts are from its Hugging Face card; Sume facts from the Image API docs. Both read 2026-10-01.

What is LLaDA-Image?

The repository's 2026-09-04 note says the Base and Turbo checkpoints were released. The card describes a unified 6B image generation and editing model under Apache 2.0, with Base at 50 steps and the distilled Turbo at 4. It lists Chinese and English text rendering and 1024x1024 generation, with height and width divisible by 16 for generation and 32 for editing.

The card's benchmark scores are the authors' own and are not repeated here.

Why is it not an API id on Sume?

Sume lists the models in GET /v1/images/models; a bare weights release is not in it. Any model value outside the catalog returns 404 model_not_found.

What do I use for edits?

GPT Image 2.5 supports reference images and an optional mask_url. References must be public HTTPS URLs; other URLs are rejected before submission.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Change the sign text to OPEN 24 HOURS, keep everything else",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/shop.jpg" } }
    ]
  }'

Should I self-host instead?

That depends on your volume and who maintains the GPUs; Sume's docs cannot answer it. If you only need an answer to "can I call it", the answer on Sume is no.

How do the two options compare at a glance?

Facts from the sections above.