Skyreels 2 I2V 1.3B with CausVid 1.3B
even faster video generation now with CausVid lora, crazily fast, 30 second step time on my GPU, download lora: https://civarchive.com/models/1585622?modelVersionId=1794411
Fast Workflow Using Skyreels 2 I2V 1.3B
This is currently my preferred workflow, aside from Wan 2.1 FP8. The main reason is that it's three times faster than the 14B variant, while offering better quality than LTX—striking a great balance between speed and performance. That makes it my go-to choice for i2v tasks.
I'm using an RTX 4080 Super, and with Wan 2.1 running optimal settings, I can generate a video in about 5 minutes without frame smoothing. However, with this workflow, I can produce the same video in just 1 minute.
Vram Requirement : 6-8gb
Prerequisites:
install custom nodes:
https://github.com/yolain/ComfyUI-Easy-Use (optional, you can just replace the node for generator, not really important)
download model from: https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Skyreels
select Wan2_1-SkyReels-V2-DF-1_3B-540P_fp32.safetensors
adjust settings for better quality:
resolutions
step count
teacache
wan video slg
Description
FAQ
Comments (12)
So Skyreel 1.3b is better than Wan 1.3b ?
wan 1.3b dont have i2v, here's how the skyreels shine here : )
@Cyberfolk Oh indeed, well your workflow didn't worked at first, but after removing torch compile and changing randomly some stuff it worked (not as fast as your original version but still fast)
@fouchardmilcoupes311 oh ya there is torch compile, sage attention, and fp16_fast, for extra speed
@fouchardmilcoupes311 correction: there is a wan1.3b i2v using VACE
wan = 16 fps
skyreels 24 fps + higher max video size
why ppl not just read what developers write on there release pages...
This workflow is really easy to use. It can be used even with low video memory. It is fast and the video quality is very good. Unfortunately, the movements, such as walking, are not natural enough. I will try to adjust the parameters and continue testing. Thank you!
The accelerator, Causvid LORA, is not designed for the main video generation. But everyone is using it here because of the speed boost. So it needs to be tweaked for strength and steps when using it as a speed 'cheat'. And the ideal parameters will depend very much on the subject of your video generation. In other words there is no "one size fits all" in video generation.
Those looking for simple "living illustrations" can get away with much faster renders than those wanting ambitious animation in their vids.
@blobby99 OK
Any way for this workflow to make use of this gguf version?
https://huggingface.co/wsbagnsv1/SkyReels-V2-I2V-1.3B-540P-GGUF/tree/main
Nice, heres my result on a RTX-3080 TI FE
Max allocated memory: max_memory=6.136 GB
Max reserved memory: max_reserved=6.812 GB
Prompt executed in 45.52 seconds
I suggest to use 1.2 CFG, this way the prompt will be used better. with only 1 you have to much static/no motions. test it with a "dance" prompt.
Works - but the output goes CRAZY after the first second of video - character moving about the place like an epileptic - what's causing this?
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.