Model card: Updates section
Browse files
README.md
CHANGED
|
@@ -44,6 +44,21 @@ This repository holds the DMAD students of the paper: 4-step students of **[Mini
|
|
| 44 |
students of the **EDM ImageNet-64** teacher. The inference and training code is in the
|
| 45 |
[code repository](https://github.com/Yzmblog/DMAD) (`train/h3`, `train/wan`, `train/image`).
|
| 46 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
## MiniMax-H3 (text-to-audio-video)
|
| 48 |
|
| 49 |
Rank-128 LoRAs on the H3 transformer that turn the 50-step teacher into a 4-step generator of 1344x768 video with
|
|
|
|
| 44 |
students of the **EDM ImageNet-64** teacher. The inference and training code is in the
|
| 45 |
[code repository](https://github.com/Yzmblog/DMAD) (`train/h3`, `train/wan`, `train/image`).
|
| 46 |
|
| 47 |
+
## Updates
|
| 48 |
+
|
| 49 |
+
* **2026/10/10:** 8 GB GPUs: a 5 s [ComfyUI workflow](#ready-to-run-workflows-8-gb-gpu-and-up) for 8 GB GPUs, and
|
| 50 |
+
`--low-vram` in the [code repository](https://github.com/Yzmblog/DMAD#on-consumer-gpus) now makes 5 s videos in under
|
| 51 |
+
7 GiB of GPU memory and 15 s videos in 12 GiB 🪶
|
| 52 |
+
* **2026/10/08:** 16 GB GPUs: the ComfyUI workflows and `--low-vram` fit in 14 GiB of GPU memory, bit-identical to the
|
| 53 |
+
larger-GPU runs 🪶
|
| 54 |
+
* **2026/10/07:** Ready-to-run [ComfyUI workflows](#ready-to-run-workflows-8-gb-gpu-and-up) (4 and 8 steps, 15 s of
|
| 55 |
+
video with audio) in `minimax_h3/workflows` 🎛️
|
| 56 |
+
* **2026/10/06:** The DMAD weights of Wan2.1 (1.3B, 14B), SDXL and ImageNet-64, and the
|
| 57 |
+
[MiniMax-H3 training data](https://huggingface.co/datasets/ZhengmingYu/DMAD-H3-data) 🤗
|
| 58 |
+
* **2026/10/05:** [ComfyUI-layout LoRAs](#comfyui) of the MiniMax-H3 students, and the training code in the
|
| 59 |
+
[code repository](https://github.com/Yzmblog/DMAD) 🏋️
|
| 60 |
+
* **2026/10/02:** The 4-step MiniMax-H3 students (`lora_critic`, `full_critic`) released 🚀
|
| 61 |
+
|
| 62 |
## MiniMax-H3 (text-to-audio-video)
|
| 63 |
|
| 64 |
Rank-128 LoRAs on the H3 transformer that turn the 50-step teacher into a 4-step generator of 1344x768 video with
|