Ver2.01
Sorry!!! While I was reworking the node, there was a node that had been disconnected at some point. We have fixed it, please re-download again!
Ver2.00 updated contents
The addition of the ability to switch with the FLF2V model adapted to the latest version of ComfyUI, the generation of loop videos using it, and many other changes, see the blog below for more details!
English (language) “Wan2.1 FLF2V” makes it easy to create complete loop animations! | なんなこなここ
Japanese (language)【革新】完全なループアニメーションが簡単に作れる「Wan2.1 FLF2V」! | なんなこなここ
Wan2.1's advanced image to video workflow packed with a variety of features including low VRAM, high speed, long video, pixel animation, 30,60fps conversion, etc.
When using lora, it is not necessary to enter a triggerWord, it will be added automatically.
When determining the number of seconds, there is no need to calculate the number of frames; simply specify the number of seconds and the appropriate number of frames will be entered.
The default filename is, for example, if you use “florence” for the prompt
wan_(name of image used)_florence_(triggerWord loraweight)_(seed)_(year_month_day)
Generation data
Using model,lora,florence like the settings in the posted image,
A4000:vram16G "wan2.1-i2v-14b-720p-Q4_0.gguf" virtual_vram_9gb
6 seconds (97 frames) of 832x480: 20step: cfg6: 1403.17s: (vram usage 60%)
6 seconds (97 frames) of 1152 x 480: 20step: cfg6: 2398.84s: (vram usage 75%)
7 seconds (113 frames) of 1152 x 480: 20step: cfg6: 2929.62s: (vram usage 85%)
4 seconds (65 frames) at 1280x720: 20step: cfg6: 2805.12s: (vram usage 83%)
5 seconds (81 frames) at 1280x720: 20step: cfg6: out of memory
A4000:vram16G "wan2.1-flf2v-14b-720p-Q4_0.gguf" virtual_vram_9gb
6 seconds (97 frames) of 512 x 512: 20step: cfg6: 1219.25s: (vram usage 86%)
6 seconds (97 frames) of 832 x 480: 20step: cfg6: 2074.77s: (vram usage 60%)
6 seconds (97 frames) of 1152 x 480: 20step: cfg6: 3427.09s: (vram usage 76%)
3 seconds (49 frames) of 1280 x 720: 20step: cfg6: 2834.66: (vram usage 70%)
See the following blog for a description of this workflow (Ver1.00)
English language Wan2.1 advanced workflow ncluding low VRAM speed-up to 60fps
Japanese language 【Wan2.1】低VRAM高速化60fps化などの高機能ワークフロー
Description
FAQ
Comments (4)
I have to know, your preview vids, how long did they take to make? Also how much vram do you have? It would give me a good estimate how long it would take on my machine as well as I'm, sure others that checkout your post
I see, I don't have a PC with GPU, so I usually use A4000 with vram16g from paperspace, and I added the current generated data in the Description.
@mercedes RAM thrashing is the curse of local video generation. One wants linear time on a home GPU- which means double the frames double the time, or double the rez, double the time. When RAM thrashing kicks in, render times go wild. The trick to linear time is keeping only intermediate data in VRAM- but most workflows waste VRAM keeping model data etc loaded.
@blobby99 ComfyUI offloads loaded models to RAM, but from my research, it seems that there is currently no way inside comfyui to freely clear model data offloaded to RAM, so the only way is to terminate unnecessary tasks or add RAM.
