Ant's LLaDA-Image puts diffusion on both ends of the image stack
- Source
- x.com
- Date
👋 LLaDA-Image combines the LLaDA2.0-mini dLLM understanding backbone with a 6B DiT for high-quality image generation and instruction-guided editing. Both backone and Image-Gen are diffusion models, trained in a unified framework. 🔑 Image-first training at scale: Of ~220M cumulative generation-training samples, >90% use image-only supervision. Image-text pairs are introduced later for language alignment. The result is a model family that covers high-quality generation, instruction-guided editing, and fast 2-4-step inference with LLaDA-Image-Turbo. #OpenSource #inclusionAI #dLLM
Context
InclusionAI's open-source release LLaDA-Image, published through its ant_oss account, pairs a diffusion-based , LLaDA2.0-mini, with a 6-billion-parameter diffusion image generator, training both within one unified framework rather than combining a standard autoregressive language model with a separate image-diffusion model.
The company reports that of roughly 220 million cumulative training samples, more than 90 percent used image-only supervision, with image-text pairs introduced later to align the model with language. A "Turbo" variant does instruction-guided image editing in 2 to 4 sampling steps rather than the many steps typical diffusion image generators require.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
- alignment — The work of making AI systems actually pursue what their builders and users intend, rather than something subtly or dangerously different.
Checking sign-in…
Loading comments…