SVI Extend
https://github.com/vita-epfl/Stable-Video-Infinity/tree/svi_wan22
Create videos and extend them seemlessly using SVI.
Following SVI LoRAs are mandatory:
switch between default behaviour, anchor_samples and end_frames within the same subgraphs
connect an image to a part and enable the respective toggles to use end_frames or anchor_samples
NEW! v3
Extend existing videos using https://github.com/wallen0322/ComfyUI-Wan22FMLF
enable "video extension" toggle inside the settings
uses source video resolution by default
rescale video using the megapixel slider by enabling "video rescale" toggle
use included version of the nodes from inside .zip or download the latest version straight from the git if issues arise
More info inside the workflow.
AIO i2v+t2v
All in One workflow for for basic WAN 2.2 video generation.
Following features included:
Switch seemlessly between 2 and 3 sampler solutions
Toggle between i2v or t2v
Postprod
Facedetailer
uses t2v Model + LoRA for inpainting - resources needed included in workflow
Toggle between GIMM VFI and RIFE VFI Interpolation
Upscale
Tensorrt Upscale with Model
Basic Video Upscale with Model
RTX Video Super Resolution Upscale (insanely fast for decent quality)
Frame Clipper
Seamless Loops using custom RIFE nodes https://github.com/Artificial-Sweetener/comfyui-WhiteRabbit
Upscale + Interpolate
I recommend using this workflow instead of upscaling with the generating workflows, since you never really know what kind of results you get, ending up upscaling a bad video and wasting time. I included toggles so you can't use multiple interpolation or upscale nodes at once by mistake.
This includes:
WAN Facedetailer
use any WAN 2.2 T2V low model + the following T2V LoRA:
lower resolution from 768 to 512 if you have VRAM issues
Put the following file into "ComfyUI\models\ultralytics\bbox":
WAN Refiner (massive VRAM cost)
increase denoise if you want more inpainting
Sharpen, Gamma, Brightness and Contrast control
Frame clipper (remove unwanted frames at the start and/or end)
GIMM VFI + RIFE VFI interpolation (I recommend GIMM VFI, much higher quality but also much slower)
Tensorrt Upscale + Basic Video Upscale
both use basic image upscaling models
Tensorrt (faster than Basic Video Upscale) with AnimeSharp4x is recommended for anime
RTX Video Super Resolution Upscale
insanely fast
decent quality
FlashSVR + SeedVR2
experimental
Video upscale models that are more intricate than basic image upscaling models
haven't had great results for anime yet
takes a LOT longer
Saving last frame for manual extensions
mmaudio
added Audio combine node
combine audio from an existing video with the generated audio on top
generate nsfw audio with the nsfw model and then combine that video with another generated audio track from the base model for background noises
removed interpolation for easier and faster audio generation - you have the following options:
upload raw unupscaled video to MMAudio Video node and upscaled video to Combine video node
upload upscaled video to both nodes but lower custom_width and custom_height of the MMAudio video node to about half for faster generation and to prevent VRAM issues
upload raw video to both nodes and upscale afterwards
Inspired by https://civarchive.com/models/2137833
Following resources necessary (ComfyUI\models\mmaudio):
https://huggingface.co/Kijai/MMAudio_safetensors/resolve/main/mmaudio_synchformer_fp16.safetensors
Description
added FLF2V nodes (EXPERIMENTAL) https://github.com/Well-Made/ComfyUI-Wan-SVI2Pro-FLF
You can mix and match them with other SVI Extension subgraphs
has the same issues as native FLF2V with endframe flashing if timescale is not correct or when using lightning LoRAs. Haven't found a consistent way to fix it
not a huge issue if you plan to only use them at the end for looping since you can just trim the last 5 frames
added WAN Motion Scale https://github.com/shootthesound/comfyUI-LongLook
Can speed up motion
values around 1.2 are recommended
can introduce artifacting
2 different Workflows
native Samplers
ClownsharKSamplers with much more granular control
FAQ
Comments (20)
Another discovery:
- the upscale from the AIO workflow may introduce eyes artifacts/flickering - I noticed this by comparing the normal raw video with the upscaled one.
- better to use the "Upscale+Interpolate" workflow for any upscale at all.
Edit: Or if you don't want to use a separate workflow for upscaling and interpolation ("Upscale+Interpolate"), activate the Facedetailer in the AIO one. It repairs the eyes for good.
The artifacting is actually from the RIFE interpolation. It's not as stable as GIMM VFI but GIMM takes a lot longer since you need to interpolate first and upscale afterwards because GIMM VFI doesn't support resolutions higher than 1280.
I generally don't recommend doing postprod in the same workflow since the video output can still be very unpredictable, batch generating raw first and then upscaling in another WF makes more sense most of the time.
@LastAssignment Holy moly, this makes sense now... I thought it's from the upscaler. Good to know in the future.
But I still remember even RIFE was giving good results since like 1-2 months ago on other workflows, Idk what happened since then, maybe ComfyUI updates broke something from the comfy-frame-interpolation custom node or somethin'. Anyways the author of that node didn't update anything in almost a year, so it could make sense.
Edit: I'm always using cloud GPU services (renting GPUs) for WAN 2.2, so I'm always using the latest comfyui commit (there's no way to use an older one, because the comfyui manager will refuse to install any custom node saying "warning: update comfyi to the latest version"). I tried to modify the comfyui manager by deactivating the safety checks and everything, but it still didn't work, so yea - I always have to use the latest comfy commit version.
@TekeshiX Oh, yeah that's definitely annoying. I try to test a version first before switching over completely, atm I'm running 0.12.3, will test the newest version in the weekend.
@LastAssignment, I see you're pretty experienced with all this video generation stuff with wan.
In your experience, increasing the low/high steps even if using lightning LoRAs will give a much better quality overall?
Let's say 6 steps HIGH and 12 steps LOW even if for the Lightning (Lightx2v) loras 4-8 steps in total are sufficient.
@TekeshiX Diminishing returns hit pretty early on with lightning LoRAs at around 6 high + 6 low which is the farthest you should go to see improvements. It's always dependent on the other LoRAs, seed and starting frame, there's no real hardcap. The biggest difference in stepcount is noticable when using 3 samplers to circumvent the downside of muted motion from the high lightning LoRA by using a "base sampler". This sampler has no lightning LoRA and uses >3.5cfg for a couple of steps and then you pipe that foundation into the normal high+low lightning LoRA samplers to make sense of it.
If you want to go for the highest quality possible with relatively low steps then I recommend trying the TripleKSampler with automatic base step calculation and setting the switch_strategy to "i2V boundary (refined)" https://github.com/VraethrDalkr/ComfyUI-TripleKSampler
I do use a similar approach in my WFs just with manual base steps to try to generate them quicker, since that TripleKSampler needs around ~5 base steps (take double the time of normal steps since >1.0 cfg double generation time) + the additional 6-8 lightning steps to be worth it.
@LastAssignment, thanks, man! Appreciate the explanation. One last curiosity: are you using a 5090 or 4090 for your generations?
@TekeshiX I'm actually using a 5080 + 96GB RAM. Comfyui is pretty good at memory management, it allows to load even FP16 models straight out of RAM without any additional VRAM cost
This guy added something that trim those bad frames on loops automatically (FLF2V). I cut the last 3-5 frames from videos generated with your workflow manually in adobe premiere.
Maybe you're interested: https://civitai.com/models/2424922?modelVersionId=2728138&dialog=commentThread&commentId=1124648
@TekeshiX Yea, just saw he added that latent cut node, I'll try it out. I'll probably rework the SVI WF again because I'm tired of going into the subgraphs constantly to change individual part loras, lmao
@LastAssignment Hahaha, okay. Still using AIO more. I mean I don't really find why I should use SVI when I can continuosly use FLF instead and basically do what SVI does but a lot better and with more motion. Heard some people saying that SVI 2 has some downsides as like bad motion, weaker prompt adherence...
@TekeshiX The transitions are just much better using SVI, I haven't had much motion issues using 3 samplers yet either. Transitions will always look a bit choppy using basic first last frame compared to the more organic SVI approach by using the motion from the last latent. Not to mention the last frame stuff for SVI can fix the motion too if you need more drastic transitions.
@LastAssignment Mhmm, okay. Guess I'll give that a try.
@LastAssignment Also if you'll rework the SVI workflow, the addition of the Facedetailer option to it would be really nice.
Thanks and keep up the good work!
Well done! Gonna try the newest workflow asap!
Installation Error: Installation failed: comfyui-customnodes-template@latest
How I can fix this?
You don't need to install them, I don't know why ComfyUI manager says they are used in the workflow but I don't have them installed either and it works just fine.
@LastAssignment Okay thank you for answer