Smittie's SVI2 1080p 60 FPS Workflow
Description
With this workflow you can generate a longer video in 1080p and 60 FPS without color or motion mismatch between the video segments (or choose cross-fade or jump-cut as transition) and inject anchor images (not end images) as further detailed guidance. You can also continue videos of ANY length (series, movies, etc.) without running into RAM issues.
If you just want to use I2V or continue videos that are only a few seconds long, then you can still use workflow v4. This one doesn't require the ComfyUI-Terminal node.
At least 16 GB of VRAM recommended, but with more quantized versions of the Wan 2.2 I2V diffusion models you can go even lower. This workflow was tested with 24 GB of VRAM and 64 GB of RAM.
Install Instructions
Those are specific to the workflow version. Download a workflow and do what the "Initial Setup" note says.
Video Instructions
If you don't know what to do, you can watch the Stable Video Infinity Tutorial by AI Search. He works with a different workflow, which I have built upon.
Acknowledgement
Most credits goes to the Stable Video Infinity, Lightning, Wan, ffmpeg, ComfyUI and CivitAI team, as well as kijai, AI Search, darksidewalker, Joviex, GACLove, jeankassio and Firetheft. Thank you guys!
Description
added bool to easily switch between I2V and Continue Video; no bypassing and re-linking required anymore
reduced RAM usage for continue video
added preview for each video generation segment
added anchor image for each video generation segment that can be used as a detailed guidance; it is not an end image, see workflow notes for more details
updated notes
FAQ
Comments (15)
Thank you very much for updating your excellent workflow. But for some reason, the three open segments give the same video without adding new fragments. Everything worked fine in the previous versions. I kindly request that you post a diagram with at least three open segments. It's a bit complicated at the moment.
So you were able to download v3? (I ask as I still thought my new version was shadow banned as the example video and image was blocked by the automatic review: said it is not AI and NSFW)
Thanks for the feedback. You can enable those easily yourself. I changed it that the standard is now only one active and the preview for it. It won't be saved immediately anymore to save space and time. You have to enable that in "7. Save Video", if you think your video looks good in the preview.
Read the note above the first preview, how I think the workflow is used properly.
Ohh and don't forget to use the new bool to switch between I2V and continue video.
This workflow works great, but I'm having a hard time understanding the proper way to prompt it. There is a string multiline text box, but it affects all of the samplers, and I don't know if there is a special way to write the prompt so that the scene is played in a sequence? And I don't know if there is an easy way to write separate prompts into each sampler, because it seems that now you have to disconnect the "get text prompt" node from every sampler in order to be able to write in it's text box?
Thank you for the feedback. As the note above the string multi line says, it is only used for all the segments that you don't want to set an individual prompt. It is useful, if you either have a infinite camera dolly forward movement or some repeating element that most segments should use.
But yeah, unfortunately, to write an individual prompt, you have to disconnect the line as you said. I think I will streamline that with the next update.
@Smittie thanks for the clarification. I have one more question - i'm using DaSi checkpoints but I'd like to try native I2V A14B fp8 scaled models, without lightning loras, and i always get blurry results, regardless of the various text encoders I've tried. I've also increased the number of steps but no go. Maybe i'm missing something. Could you point out the exact settings and encoders that work with the native model ?
@loucypher83827 For native non-lighting models I think the correct settings were:
ModelSamplingSD3 shift - 5 or 8
BasicScheduler Steps - 20 or 30
SplitSigmas Step - half of BasicScheduler steps, so 10 or 15
CFG - 3.5
But if that does not work, it could be that the SVI LoRA, that is needed for the long consistency, isn't supposed to work with native non-lighting models.
Thanks for the workflow! The detailed notes really helped get good quality vids out of it from the get go and not waste a day on experimenting.
Thank you for the positive feedback. I am glad that my workflow and notes are helpful.
This is a great workflow -- my go to for any long length vids! Thank you.
Thanks for the positive feedback!
very cool, I got very confused with the 3B section, so I bypassed everything there, and also had to bypass the lazy switch at 7, but after that, everything worked... I'm guessing 3B is to continue a video already made somewhere else?
Yes, 3B is to continue a video you already have on your HDD/SSD
With the next update I will streamline the workflow, that it looks not that complicated anymore.
But you can keep 3B active and the lazy switch too. Just change the bool in the settings and you're fine.
@Smittie I mean, i had it switched to false, so it would go to I2V instead of V2V, but the workflow would stop after generating the sample image, things worked only after I bypassed all that, so dunno what I could be doing wrong.. thanks for the reply, and looking forward to the update!
@artificialotaku mmmh... perhaps the Lazy Switch still tries to evaluate the other components, even though the boolean should not lead to it? You can try to add a video to 3B and don't bypass it. Keep the bool "IsContinuingVideo" as false as is to use I2V.
If my theory is correct, it will then evaluate the video, but not using it further.
@Smittie oh, I get it, so the workflow needs a video there, even though you may not need it? I'll try that and will let you know how that went