whatmodel.sh / video / Open-Sora Plan v1.3 (I2V)
Open-Sora Plan v1.3 (I2V)
↗ share
+ contribute
Open-Sora Plan v1.3 (I2V) is an open-weights video model from Peking University (PKU-YuanGroup), released October 2024. Its published size is 2.78B parameters. Licence: mit.
Model information sources
Image-to-video diffusion transformer only: OUR COUNT from the safetensors headers of the official any93x640x640_i2v shards (pinned revision): 2,495,151,360 + 287,562,272 = 2,782,713,632. The HF API gives no total for this multi-model repo. Text encoder, WF-VAE and the prompt refiner are not included; the T2V transformer (any93x640x640) is a separate 2.77B checkpoint. Observed 2026-10-07.
$ how it compares
The video models ranked either side of Open-Sora Plan v1.3 (I2V), out of 139 with a score, then more from Peking University (PKU-YuanGroup).
| # | model | score | where | ||
|---|---|---|---|---|---|
| 100 |  Mochi 1 | 40±2 | api price unknown +local | ||
| 101 |  Runway Gen-3 Alpha | 40±3 | comparison only | ||
| 102 | Open-Sora Plan v1.3 (I2V) | 39±7 | local | ||
| 103 |  UniVideo | 39±3 | comparison only | ||
| 104 |  T2V-Turbo (MS) | 39±6 | comparison only | ||
| more from Peking University (PKU-YuanGroup) | |||||
| 71 | Open-Sora Plan v1.3 (T2V) | 49±7 | local | ||
$ what it's good at
other tests1 not in a task · 81% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WorldModelBench (human total) | 6.62 / 10 ✎ | ▅··▅▃···▃··█ beats 3/8 | 81% |
$ where to run it
Other ways to run it
- Run it as published · needs ~24 GB, NVIDIA only · from github.com/PKU-YuanGroup (official)
git clone https://github.com/PKU-YuanGroup/Open-Sora-Plan && cd Open-Sora-Plan && git checkout v1.3.1 && pip install -e . && hf download LanguageBind/Open-Sora-Plan-v1.3.0 --include 'any93x640x640_i2v/*' 'vae/*' --local-dir osp13 && hf download google/mt5-xxl --local-dir mt5-xxl && torchrun --nnodes=1 --nproc_per_node 1 -m opensora.sample.sample --model_type inpaint --model_path osp13/any93x640x640_i2v --version v1_3 --num_frames 93 --height 352 --width 640 --crop_for_hw --text_encoder_name_1 mt5-xxl --text_prompt examples/cond_prompt.txt --conditional_pixel_values_path examples/cond_pix_path.txt --ae WFVAEModel_D8_4x8x8 --ae_path osp13/vae --save_img_path ./samples --fps 18 --guidance_scale 7.5 --num_sampling_steps 100 --max_sequence_length 512 --sample_method EulerAncestralDiscrete --seed 1234 --num_samples_per_prompt 1 --rescale_betas_zero_snr --prediction_type v_prediction --save_memory
Weights and documentation: huggingface.co/LanguageBind ↗
$ how the score was worked out · WorldModelBench: Judging Vide… 81%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 WorldModelBench: Judging Video Generation Models As World Models | 1 | 81% |
| prior · pull toward the middle | 19% | |
| weighted average | = 39.3 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json