Cinematic Hard Cut - CivArchive (CivitAI Archive)

Cinematic Hard Cut - v3.0 I2V High Noise Wan

NSFW

V3 update: Included smash cut type transitions. "The camera makes a smash cut..."

As much as I love Wan, if it's one thing it's not good at, it's cuts.

This lora is an experiment in greater shot control, by being trained on hard cuts. Most of the clips are also cut on action, fwiw, to allow for a continuous flow.

It's captioned with "wide-angle", "mid-shot" and "close-up". But I think Wan already handles them well.

Prompt format:

[a brief description of your initial shot.]

the camera makes a hard cut to [resulting shot]

[what happens after]

You can string several cuts. If you do, it works best if you also change the type of shot.

Where's the low noise model? There isn't one. It seems to work well with only the high noise model. I'll still train and test a low noise model, to see if it will improve consistency during cuts.

Play with strength if the scene changes too much. Too low strength will make it a dissolve transition (at least testing seems to imply that).

Using lora's (for the target cut) you can get pretty far.

If the cut doesn't 'take':

Try using less other lora's.
If you use a high resolution, lower it.
Try generating a shorter clip.

Description

including 'smash cut' type transitions

FAQ

Comments (24)

CobaltRiftDec 30, 2025

CivitAI

It looks better than many sequences I’ve seen, however her staff changes hands which disrupts continuity.

neph1

Author

Dec 31, 2025· 2 reactions

I guess they made a mistake in the editing room. It happens. ¯\_(ツ)_/¯

Kunio_114_Dec 31, 2025

CivitAI

このLORAをありがとう。これがあれば私は何だってできる気がする。

Kunio_114_Dec 31, 2025

CivitAI

続けてすみません。すごい発見をしました。それは、トリガーの前と後の文章量の違いが、そのままカット位置につながることが多いってことです。これはすごく便利なコツです。

rsrss285Jan 9, 2026

CivitAI

What's the difference between the smash cut lora, and this hard cut lora?

neph1

Author

Jan 9, 2026· 2 reactions

They used to be separate, but from v3, I've included the best training clips from smash cut into this one. It was simply better to begin with. I will probably do the same with quick cuts for v4.

billysaltzman625Jan 11, 2026· 1 reaction

CivitAI

Should this be in both high noise and low noise?

neph1

Author

Jan 11, 2026· 1 reaction

just for high noise.

murazisunJan 15, 2026

CivitAI

The cut shot in the example is very simple, but my cut shot is a lens shift on the same screen. I also used standard words, so I am very curious, is it a problem with my original picture, or is it a grammatical structure? Every time I cut the lens, I always move the lens to transition and pan to lock it, instead of changing the angle of view to focus on the target. What is the point in the original picture? Can you provide the original image of the sample video? If I generate a vertical screen ratio, will there be any problem with using horizontal images? If there are two people talking in the vertical screen, the two people will easily overlap in the vertical screen.thks

neph1

Author

Jan 15, 2026

Which video do you mean? In all cases of the example videos, the image you see at the beginning is the start image. One thing I've noticed is that it usually works better if you change the type of shot, ie "close-up" to "mid-shot", or "wide-angle". Same to same limits camera movement, even if you tell it change subject, for example. This is probably due to a lack of training examples. It might be the same in your case, that it's not trained on the type of change you want to do. If you have any examples, I could try to incorporate some the next time I update it. Also, if you're using other loras, try to lower their strength.

tipwakaJan 23, 2026

the camera makes a hard cut to｛Close-up description}.
Always smooth transitions，strength:1~1.2

Kuminer69Feb 7, 2026· 3 reactions

CivitAI

Works good for starter images with Subject overlaying a Background. Then Camera Makes a Hard Cut to Subject interacting with the Background. I actually stopped using Qwen Image Edit and use this Lora instead.

Psy_pmpFeb 27, 2026

CivitAI

I couldn't find a single model that this lora would work with. I only tested it with SVI

ElGordoAIMar 4, 2026

CivitAI

this works EXTREMELY well with nsfw :)

afvfygt00o491Mar 31, 2026

与wan2.2呢？

PlasmadoseMar 8, 2026

CivitAI

Who's got the exact magic words needed to keep the exact same people and same body types after the hard cut? I prompt as such and either get different people and/or almost always different body proportions.

150591822982Mar 26, 2026

CivitAI

thx for the share,work fine.
how to add more cut,i add 5 cut but it seem just do 3

neph1

Author

Mar 26, 2026

It's not really trained on multiple cuts (will add for next version). https://civitai.com/models/2113025/cinematic-fast-cutting-previously-quick-cuts is trained on up to 5 cuts. It might do better, or in combination. In general, Hard Cut does multiple cuts better, though.

teeay25225Apr 30, 2026

CivitAI

Two questions: Does it changes faces? And what strength to use? Thanks in advance

AlbanianDoorknobMay 3, 2026

If you put the strength to something like 8 or 30 (if using comfyui) then you should be able to get the multiple faces that you're requesting but it will only generate a file and not a live broadcast of changes over time.

neph1

Author

May 5, 2026

I'd say the training data is varied enough to warrant a 'no'. But I leave no guarantees. I usually start out around 0.7 and adjust it from there.

hsholmeszeroMay 9, 2026

Yes it sometimes changes for me. All Asian characters turns Caucasian at default Wan 2.2 workflow at 480p resolution, 1.0 strength. Might be more effective is strength is lowered or if resolution is higher (though slower).

seroxyk950Jun 10, 2026

CivitAI

Hello I wanted to ask you something.
I am currently also trying to train a lora with a hardcut, but for some reason, I think my lora didnt learn the hardcut transition well enough, because it tends to zoom out first and then do a hardcut or something in between.

Could you maybe share what rank/epochs you used, especially for the high expert?

Would be greatly appreciated.

neph1

Author

Jun 10, 2026· 1 reaction

i replied in more detail to your DM, but for anyone else: rank 32, lr of 3e-5, linear scheduler. short, focused clips of 33 frames

LORA

Wan Video 2.2 I2V-A14B

by neph1

Download (Beta) View on CivitAI

concept

cinematic