4 Pull requests merged by 3 people

• Add free storage API for FSDP

  #277 merged Mar 26, 2025

• Add all py files in the subfolders under the "wan" directory through pattern matching

  #271 merged Mar 24, 2025

• Add more explanation about the context parallel

  #260 merged Mar 20, 2025

• Add a pyproject and a separate INSTALL.md

  #137 merged Mar 20, 2025

35 Issues closed by 6 people

• 微调wan2.1-t2v-1.3B时，flow_shift要设置成多少比较合适呢？

  #279 closed Mar 26, 2025

• 使用FSDP加载模型，在切换模型时显存无法释放

  #263 closed Mar 26, 2025

• Is there a way to simply use the CPU

  #274 closed Mar 26, 2025

• "RuntimeError": when use T2V 14B diffusers

  #276 closed Mar 26, 2025

• Stuck in loading model

  #272 closed Mar 26, 2025

• Question about the noise schedule

  #249 closed Mar 24, 2025

• data preprocessing of train dataset

  #252 closed Mar 24, 2025

• prompt weight

  #257 closed Mar 24, 2025

• Start & end frames integration

  #264 closed Mar 24, 2025

• When will Wan-Bench be released?

  #269 closed Mar 24, 2025

• model always failing while downloading from HF

  #187 closed Mar 24, 2025

• 有一个疑惑点请教一下， self.self_attn 先在qkv 和rotary采用 sequence parallel并行， sdp 前后有两个all2all 在计算sdp时使用的是一个完整的sequence， 为什么 self.cross_attn 没有采用 sequence parallel并行， 另外 self.cross_attn的 seq 不是完整的 这样计算是不是有问题的

  #221 closed Mar 24, 2025

• 文生图模型独立出来

  #200 closed Mar 24, 2025

• models_t5_umt5-xxl-enc-bf16.pth 需要单独下载吗

  #201 closed Mar 24, 2025

• CUDA out of memory

  #205 closed Mar 24, 2025

• "Building wheel for flash_attn (setup.py)..." Loading FOREVER

  #206 closed Mar 24, 2025

• Windows Error pip install -r requirements.txt (Failed building wheel for flash_attn) setuptools.installer and fetch_build_eggs are deprecated

  #207 closed Mar 24, 2025

• Apple Silicon Support

  #208 closed Mar 24, 2025

• 图生视频突然killed

  #209 closed Mar 24, 2025

• Is the repo or any ali repo can generate image?

  #215 closed Mar 24, 2025

• fps:16

  #220 closed Mar 24, 2025

• wan2.1_i2v_720p_14B_bf16模型生成81帧的视频，耗时接近2400s

  #222 closed Mar 24, 2025

• Video generation distillation

  #223 closed Mar 24, 2025

• 2x Wan 2.1 Step by Step Tutorials for Windows and private cloud RunPod and Massed Compute for GPU Poor

  #224 closed Mar 24, 2025

• About T2V to I2V model structure

  #225 closed Mar 24, 2025

• When will the technical report be released?

  #226 closed Mar 24, 2025

• 模型能int8或者int4吗？

  #228 closed Mar 24, 2025

• Can't save the seed

  #231 closed Mar 24, 2025

• Wan 2.1 1.3B generated videos are very pixelated

  #233 closed Mar 24, 2025

• sampling steps

  #234 closed Mar 24, 2025

• The right promt and the right negative promt for the WAN

  #243 closed Mar 24, 2025

• Video

  #246 closed Mar 24, 2025

• Deepspeed stage 3

  #248 closed Mar 24, 2025

• Make me video of futball player hit the post from center

  #258 closed Mar 24, 2025

• how to timesteps in training

  #180 closed Mar 20, 2025

11 Issues opened by 11 people

• checkpoints of downstream task

  #278 opened Mar 26, 2025

• i2v-14B 随机seed生成效果差异很大

  #275 opened Mar 26, 2025

• Sage Attention vs Flash Attention Speed Comparison with Wan 2.1 - Sage Attention is 37% Faster - 720p - 14b model - tested on Windows Python VENV - no WSL

  #273 opened Mar 25, 2025

• how to generate long videos?

  #270 opened Mar 24, 2025

• Questions on the Quantization Performance in the Report

  #268 opened Mar 22, 2025

• Layernorm 设计很奇怪

  #267 opened Mar 22, 2025

• CUDA OOM with multi-GPU

  #266 opened Mar 21, 2025

• 文生视频/图生视频均失败13B

  #265 opened Mar 21, 2025

• Ollama compatible?

  #262 opened Mar 20, 2025

• Intel Processor as Co-Worker?

  #261 opened Mar 20, 2025

• Wan 2.1 VAE Reconstruction noise

  #259 opened Mar 20, 2025

18 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• I2V任务，为什么不用VAE encoder编码，而是使用新的CLIP Vision encoder进行编码

  #235 commented on Mar 20, 2025 • 0 new comments

• my one click jupyter notebook for this model

  #146 commented on Mar 20, 2025 • 0 new comments

• NVIDIA 50XX显卡的看过来

  #189 commented on Mar 21, 2025 • 0 new comments

• ComfyUI plugin for multi GPU inference

  #250 commented on Mar 21, 2025 • 0 new comments

• I2V 14B OOM solution

  #241 commented on Mar 22, 2025 • 0 new comments

• GPU 4090：使用界面 gradio/t2v_1.3B_singleGPU.py 生成一个 480*832的视频， steps 50/50后显示显存存不够用

  #52 commented on Mar 22, 2025 • 0 new comments

• 减小frame_num后报错

  #156 commented on Mar 24, 2025 • 0 new comments

• Frames per second clarification

  #15 commented on Mar 24, 2025 • 0 new comments

• Any comparisons on Vbench about 1.3B model?

  #104 commented on Mar 24, 2025 • 0 new comments

• My video suddenly turned like this. Please help me find the reason.

  #256 commented on Mar 24, 2025 • 0 new comments

• why must use flash attention

  #188 commented on Mar 24, 2025 • 0 new comments

• Question about the model's output

  #251 commented on Mar 24, 2025 • 0 new comments

• Wan2.1 Imported Failed in my ComfyUI

  #254 commented on Mar 24, 2025 • 0 new comments

• RuntimeError: FlashAttention only supports Ampere GPUs or newer

  #25 commented on Mar 25, 2025 • 0 new comments

• Using WanImageToVideoPipeline, the results doesn't work that well.

  #216 commented on Mar 25, 2025 • 0 new comments

• Request: Add Metal GPU Apple Silicon Support

  #14 commented on Mar 26, 2025 • 0 new comments

• Add macOS Compatibility

  #69 commented on Mar 23, 2025 • 0 new comments

• Update attention.py

  #116 commented on Mar 24, 2025 • 0 new comments