Tasks and inputs
Nemotron-Labs-Diffusion-Uni-8B is listed for Any-to-any, Text-to-image. It accepts text, image and produces text, image. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
NLD-Uni (Nemotron-Labs-Diffusion-Uni-8B) is NVIDIA's open diffusion-language model that both understands images and generates them, for research.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights; no per-token price is published. Your cost is the GPU time you run it on.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Unified multimodal models usually bolt an image generator onto an autoregressive chat model. NVIDIA's Nemotron-Labs-Diffusion-Uni-8B, shortened to NLD-Uni, starts from the other end: a Nemotron diffusion language model, with separate transformer paths for understanding and for generation inside one framework.
It is a research release. Treat it as a way to study the approach, not a drop-in product.
For understanding, images become continuous visual features and the model writes text. For generation, it works on discrete visual tokens, refines them through iterative diffusion and decodes them to RGB. Text can be decoded autoregressively or block by block with diffusion. Image-conditioned generation and editing are possible where the checkpoint and inference code support them.
NVIDIA validated it on a single H100 80 GB. The weights come as eight BF16 shards holding 1,557 tensors, with a hash manifest to verify downloads. The environment expects Python 3.13 and CUDA 12.6, and FlashAttention has to be built after PyTorch. Inference is plain Transformers plus the repository's model code, with an inference.py command per task. Text results serialise as JSON records and images as PNG.
Some details stay vague. The card describes the architecture in general terms. The name says 8B, but no exact parameter count is given. Context length and image resolution are left to the shipped configuration rather than a fixed limit. The licence is NVIDIA's Open Source License Agreement, so read it before building anything commercial.
The release date given is 27 October 2026, though the repository was created on 5 October.
Run the same twenty prompts through understanding and generation: describe a chart, then generate a chart in that style. You learn quickly whether the shared framework helps or whether a specialised pair still wins. For understanding-only baselines, browse image-text-to-text models. For generation, compare against text-to-image models.
Compare the model on the work you need to complete.
Nemotron-Labs-Diffusion-Uni-8B is listed for Any-to-any, Text-to-image. It accepts text, image and produces text, image. Use the publisher’s input format and preprocessing for your chosen deployment.
Choose the output size and quality your design needs. Include drafts and revisions when budgeting an accepted image.
The weights are downloadable under NVIDIA Open Source License Agreement. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Liquid AI
LFM3-Omni-2B is Liquid AI's open any-to-any model: image, video, audio and text in; text and speech out, with multimodal tool calls on-device.
NVIDIA
PixelUMM is NVIDIA's encoder-free 15B unified model that understands and generates text, images and video directly in pixel space. Non-commercial.
shanghai-ai-lab
InternLumina-U2 is Shanghai AI Lab's masked-diffusion MoE that understands images, video and 3D and generates and edits images with 8-codebook visual tokens.
sensetime
SenseNova-U1.5-8B-MoT is SenseTime's open unified model for native 4K image generation and instruction-based image editing, Apache-2.0 licensed.
Tencent
Hy-Embodied-RxBrain-1.0 is Tencent's 6.2B embodied model that reasons in text and imagines goal images, for planning, world-state prediction and video QA.
google/gemma-4-E2B-it is a source-linked Hugging Face repository assigned to Any-to-any. Its recorded task interface accepts text, image, audio, video and produces text, image, audio, video.