(LTX-2.3 Lora) - Pose Helper (I2V) - CivArchive (CivitAI Archive)

(LTX-2.3 Lora) - Pose Helper (I2V) - LTX 2.3

NSFW

(LTX-2.3 Lora) - Pose Helper (I2V)

A lora to help with animating subjects, especially in I2V.

Should improve I2V generations for letting (clothed) subjects pose and turn around.

Check out "Camera Zoom Out" lora as well: https://civarchive.com/models/2490345/ltx-23-lora-camera-zoom-out-i2v

Use "d3m0p0s3" as a prompt word
Use Lora Strength around 0.7-0.9
Preferably use with I2V on LTX-2 DEV (FP8).

Trained on 1280x704, 960x960 and 704x1280 videos. (LTX 2.3 also trained with 48fps)
Feel free to post your results, though keep it decent.
Read more below for prompt examples
Should technically also assist in animating T2V char lora's (consistency may vary)

Known issues: Objects being held in hands may disappear when turning around. It also may not be perfect but should compensate a lot for the base I2V lack of motion and lessen glitching with subjects turning around. NSFW may not fully work properly (i.e. clothes may suddenly be added) as dataset contained clothed subjects.

For prompting, use keyword "d3m0p0se", followed by a sentence describing what the subject is wearing, and whether the subject is posing or turning around.

I2V Prompt example:

d3m0p0se.
A brunette woman wearing sunglasses mini jeans and a white top is saying (not singing) seductively to the viewer "Don't you like my outfit".
She then is happily posing  on the street,  touching her clothes and hair.
Pop music plays in the background

Was trained on captions like:

d3m0p0s3, black borders, A barefoot black haired woman posing and turning around in her white flowery bikini. Details: grey floor, white wall, mirror on the right

d3m0p0s3, black borders, A white sneakers wearing brunette woman posing and turning around in her blue denim jean and top outfit with a white shirt underneath. Details: gray floor, white wall

Description

FAQ

Comments (15)

mixailckopp978Mar 19, 2026· 2 reactions

CivitAI

Wow, it's really helpful Lora

JonXL

Author

Mar 19, 2026

Thank you!

Feel free to post your results on this page ;)

brendacaroMar 22, 2026

CivitAI

I have no idea what this lora does. And in the comparison examples below I see no difference between with lora and without lora. What am I missing?

JonXL

Author

Mar 23, 2026

Hi there brenda!

Thank you for letting us know about your question.

As the description reads "Should improve I2V generations for letting (clothed) subjects pose and turn around.".

Now, LTX2.0 without additional loras would be mostly static, so people would bsrely move.

For LTX2.0 this lora made a huge difference in animating I2V bodies, hand movement, head movement, walking and turning around.

The LTX2.3 base model seems to have fixed some of the motiom issies where out of the box images in I2V mode.

Still with this lora more bodily expressions are added and there is less chance of glitching, where for example the head looks at the camera while the body is turned.

Hope this explanation helps clarify a bit!

You could also run some test yourself to find out how it improves on the base model.

FlowSpecialMar 24, 2026

CivitAI

Mm, I would say LTX 2.3 has improved IMG2VID however, I still think there's a long way. Wan 2.2 for example, came with a incredible handling of I2V, I'm hoping that LTX continue to improve in this regard. I'll say I'm certain that is going to be the case, 'cause right now is kinda cursed, ain't gonna lie.

JonXL

Author

Mar 24, 2026

I wouldn't say it's cursed, but there is definitely still room for improvement.

Just to make it a fair comparison, WAN has it's shortcomings too.

For example, when subjects are turning around, the rear is shown but there is a high chance the head remains looking at the camera. LTX has a lesser chance of this happening, and this lora even improves on it.

Last time I checked WAN also checks out at around 5 seconds of generation (at 720p) and degrades the video over time.

I think it's kinda cool that LTX decided to give us a longer length, higher resolution, audio capable, lip sync out of the box model. Sure it's not the best yet but I am not going back to WAN :)

From the LTX Discord I get the feeling that the team will be releasing more and more updates.

FlowSpecialMar 24, 2026

You're not wrong. I believe LTX is going in the right direction, and they are going to continue updating the model, I'll guess every 3 months or so, eventually, they will bypass WAN 2.2, and that's something I'm very excited for and looking forward to it. By the way, have you heard of the VBVR reasoning lora, and model for wan 2.2? It surpasses even Sora 2, or Veo 3.1. See: https://video-reason.com/bench/#leaderboard.

The explanatory video here: https://www.youtube.com/watch?v=EuEVZu2wl7g&t=257s

Please report back with your thoughts.

JonXL

Author

Mar 25, 2026

Interesting that they made a reasoning lora for Wan 2.2, looks impressive.

I wonder if they will port it to LTX 2.3 as well.

FlowSpecialMar 26, 2026· 1 reaction

@JonXL It would make a huge impact.

ArdongelApr 2, 2026

I have completely deleted WAN from my hard drive.

JonXL

Author

Apr 2, 2026· 1 reaction