Training and inference scaling
According to the model card, SANA-1.5 uses training-time and inference-time scaling, including efficient growth from 1.6B Sana-1.0 to 4.8B and depth pruning.
Open Source Model Profile · Efficient-Large-Model
SANA1.5_1.6B_1024px
SANA1.5_1.6B_1024px is a text-to-image model from Efficient-Large-Model. According to the model card, it is a 1.6B-parameter linear-diffusion model for 1024px generation.
- Publisher
- Efficient-Large-Model
- Task
- text-to-image
- Model type
- Unknown
- License
- apache-2.0
- Library
- sana
- Publication status
- Accepted · not indexed
Model overview
SANA1.5_1.6B_1024px is published by Efficient-Large-Model as a text-to-image model. According to the model card, developed by NVIDIA and Sana, it is a 1.6B-parameter scalable linear-diffusion-transformer model for 1024px-based images.
Recorded capabilities
1024px BF16 configuration
According to the model card, the publisher lists 1.6B parameters, torch.bfloat16 precision, and 1024px-based multi-scale generation.
Published source repository
According to the model card, source code is available at the NVlabs Sana GitHub repository, recommended for training and inference with Flow-DPM-Solver.
Research-use scope
According to the model card, direct use is for research including artworks and design, while factual representations of people or events are out of scope.
Use cases in the source record
- 1024px text-to-image generation with the scalable linear-diffusion-transformer design described in the card.
- Research generation of artworks and design-process imagery, as stated in the card's direct-use description.
Limitations and unknowns
- No weight-derived Safetensors parameter count was extracted; the 1.6B figure is a publisher-reported card claim.
- No independently measured evaluation results were extracted; GenEval and DPG-Bench mentions are publisher-reported claims without extracted scores.
- According to the model card, the model was not trained to be factual or true representations of people or events.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: Efficient-Large-Model/SANA1.5_1.6B_1024px
Captured: Unknown. Processed: 2026-09-07T19:35:05.034353+00:00.
🐱 Sana Model Card Model We introduce SANA-1.5 ，an efficient model with scaling of training-time and inference time techniques. SANA-1.5 delivers: efficient model growth from 1.6B Sana-1.0 model to 4.8B, achieving similar or better performance than training from scratch and saving 60% training cost; efficient model depth pruning , slimming any model size as you want; powerful VLM selection based inference scaling , smaller model+inference scaling > larger model; Top-notch GenEval & DPGBench results. Detailed results are shown in the below table. Source code is available at https://github.com/NVlabs/Sana . Model Description Developed…
F001F002F003F004F005F006F007F008F009F010F011F013F015F016