#6767th of 70 in Open models
show-o
Unified transformer for multimodal understanding and image generation
- Stars
- 2k
- License
- Apache-2.0
- Last commit
- Jan 2026
- Language
- Python
1,902 on 21 Apr 20261,986 on 10 Nov 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
Show-o and Show-o2 are research models from NUS Show Lab and ByteDance that use one transformer for multimodal understanding and generation. Show-o (1.3B) handles captioning, VQA, text-to-image, inpainting and extrapolation, mixing autoregressive text with discrete diffusion on image tokens. Show-o2 (1.5B and 7B) works in a 3D causal VAE space, adds video understanding, and uses autoregressive and flow-matching heads.
Who it is for: Researchers and ML engineers working on unified multimodal models
Strengths
- One model covers VQA, captioning, text-to-image, inpainting and extrapolation
- Checkpoints on Hugging Face for Show-o (256 and 512 res) and Show-o2 (1.5B, 7B)
- Training code for pre-training and instruction tuning, with accelerate and DeepSpeed configs
- Hugging Face Space demo available; Apache-2.0 license
Weaknesses
- Research code driven by CLI scripts and YAML configs; no server or UI
- Stage 1 training needs manual edits because RefinedWeb loading uses internal packages
- Inference demos log results to wandb, which requires a login
- Mixed-modal generation and visual tokenizer training are listed as pending features
What it needs
- Needs PyTorch, wandb, accelerate, Hugging Face Hub
- Models: Show-o 1.3B, Show-o2 1.5B, Show-o2 7B
Quick answers
Is show-o open source?
Yes. Its code is on GitHub under the Apache-2.0 license.
Also in Open models
See all 70| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier for typed choice, score and yes/no decisions | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 79 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | yolov5PyTorch YOLOv5 object detection, segmentation and classification models with training code | 76 out of 100 |