Just ran a quick side-by-side, with telling results.
LTX 2.5:
Silhouette Matte ML + Matte Assist + Matte Refine:
Flame AutoMatte
Source clip: https://artlist.io/stock-footage/search?terms=697723
This test was done at 3200x1900 resolution, for 49 frames.
Flame ran with it’s usual parameters on Linux.
Silhouette ran on Linux, used about 10GB NVRAM
LTX ran on Linux, used about 90GB of NVRAM, using bf16 version of LTX 2.5 (42GB model file)
Flame results were almost instantenous after the ML model loading
Silhouette took about 1min
LTX ran in about 2.5min
The LTX model was a more powerful version than what Flame or Autodesk distribute. So it had the advantage.
LTX model has some frame count limitations, so running it on long and large clips will take a bit of extra work. Also on long/large clips you run into RAM limitations of Comfy itself.
I had existing LTX workflows for Comfy, so setup was less than 10min.
All workflows read/wrote EXR sequences, only final export for posting here was .mp4
Is LTX 2.5 using same process as VideoMaMa: Mask-Guided Video Matting via Generative Prior, or something else? I was able to find this: “The IC-LoRA conditions on the latents of the RGB reference video and generates the corresponding alpha matte as a video, with an empty text prompt. Because the reference stays attached for the entire denoise (stage-1-only inference at native resolution), the matte stays aligned with the source down to fine edges.”
I don’t know the exact mechanics of LTX. But it is one core model and then a number IC loras for different workflows, different from the earlier loras which just bias style.
And the prompt was empty indeed on my test. Works like AutoMatte in finding the dominant object in frame.
It’s worth noting the business model here. LTX is not like Qwen or Wan which have been open models under GPL or similar license, until their owners think they had enough PR and lock future version up behind paywalls as happened with Wan 3.0.
LTX is free for commercial use under a certain turnover threshold. Above that you have to pay for a commercial license. The company’s goal is to make models, they’re not a byproduct of something else. I hope there will be more like that.
Not that paying reasonable API rates is a problem. As long as we have a local option.
Essentially the Nuke Indie model.
ADSK needs to stop spending resources making in-house sub-standard models, and focus on seamless integration of 3rd party SOTA models. Batch should have been what ComfyUI now is.
You say that autodesk should have turned batch into comfyui, but think how grubby the pearls would have become from all of the fervent clutching.
Reflect on just how vibrant and productive the community development has become.
Years of domain specific knowledge and insight has brought about significant advances, backed up by twenty years of Reddit generality and an electricity bill you didn’t have to pay for.
I’m happy that elbows are finally pythonic.
It helps organize the feeds into the comfyui batch nodes.
I see. Thanks for the info. I’ll have to look more into it.
Some possible pointers on why this model performed so well, and also implicit limitations.
The first one is that is appears to attention over the whole clip, rather than frame by frame. That helps it deal with frames where conditions from frame-by-frame models get worse because of contrast and color changes. It also means it can handle occlusions better. Temporal stability is inherent to its design.
The downside is that this requires significantly more compute and doesn’t scale with clip length. The model is currently limited to 145 frames, and because of whole clip attention you may have limited luck breaking longer clips in to segments without getting noticeable issues at the seams. May have to pick strategic seams.
The other factor is - classic loras used to just nudges the models weights to favor a specific trained preference. An IC lora adds the additional reference input (IC = in context), and with it additional training data.
This has two consequences in this particular model: this model by-passes the hallucination phase of the model processing. The matte pixels you see come from the reference input, they’re not pixels the model has hallucinated, as all the other models do we are familiar with.
Other matting models decide for each pixel, what matte should look like based on their training. Hence the flickering and inconsistency. This model just decides if a pixel is foreground or background based on its training data. Foreground = output is max(r,g,b) for all 3 channels. Background is black.
The second consequence is that the core model is a 22B diffusion model, which has seen several magnitudes more training data than all the matting models ADSK and BorisFX have trained. This is at the scale of other diffusion models, not matting aux models. That’s an unfair advantage towards the training capabilities of the other models.
And in my case I ran the bf16 version of the model, not the further quantized smaller variants. So it had additional precision to work with.
So a lot more memory, a whole lot more compute, and some stricter boundaries on what can be processed. But if you stay within those boundaries, holy cow, this really works.
So this model does take a whole different approach to ML matting than what we’ve seen until now. Similar to how SmartRoto has taken a whole different approach by pursuing ML-masking instead of ML-matting.
A question for ADSK is that currently they’re shipping models that are tuned to a wide range of GPUs. Which is nice if you have a A5000 ADA card and get nice automatte results (maybe not from fine hair, but a wide range of objects - we’ve seen the kudos to AutoMatte).
But it also means that if someone goes out and sources a bigger GPU, that could handle a lot more compute, they’re not getting better results for the most part. It’s one model for all. There is a point to be made that there should the S, M, and XL models.
If you need to crack a hard nut and you have the compute, I don’t mind downloading a 42GB model file and having it go for it. And I doubt any of you would.
Indeed. Pretty much everything you said is more or less what I found as well from research. Off course the question of where all these models are trained and how also remains somewhat vague or not vague at all, depending on how you look at it.
I guess there is no free launch anyway you look at it.
Another video release , in Portuguese. CC on. or auto translate if you don’t speak the language.
Someone in another Discord had a hard time with a hand motion to work in LTX 2.5.
I thought it might be interesting to see some more scenarios. The original test was about fine hair details. So here are 7 different hand movement scenarios based on ArtlistIO clips.
In all cases rendered out clip as EXR Openclip out of Flame, my custom LTX workflow, reimported matte as EXR Openclip and comped colour source as matte overlay.
For most I tried to run them as 1920x1080, but in some cases they slipped through at higher res. And all clips except one were rendered at 70 frames.
518431 ran at UHD for full 145 frames. That took 39min to complete. The runtime of the rest, especially the HD clips was much shorter. Many finished in 64s, and one came in at 11min.
There was some cropping needed on the mask since LTX runs in multiple of 32 in resolution. So your mask comes back as 1920x1088. Resize with edges cropped.
If you look closely most have minor defects or some dilation that may need tweaking. But they’re still amazingly clean for what it is.
The numbers under each video is the ArtListIO asset id, if you want to download and test for yourself.
518431
603805
6587146
6614181
256716
762582
766566
I just did a test comparing it with SAMMIE. A challenging shot: fast movement, motion blur, backlighting, hair, etc.
Overall, I find SAMMIE more consistent, with better masking and better consistency in human’s shape than LTX. LTX may have better results when it comes to hair. It might be possible to combine both approaches to get the best result.
So, I still prefer SAMMIE, mainly because of how easy it is to install and use. LTX requires around 80 GB of models, and using ComfyUI and its workflows comes with a pretty steep learning curve.
I think that will be true for a lot of these. You will need more than one tool, and in some cases you need to combine mattes from two of them.
This level of effort is worth it when traditional methods of keying and roto fall flat or are more expensive than the time spent here.
Thanks for the post and the work!
This is pretty impressive. Enough so that I will add it to my AI-Matchbox nodes (BCE). I already have a SAM3 node, but this seems like it would be a good addition, as more tools gets us more results.
Tough to run in the cloud, tho. comfy.org has a custom build API now, maybe I can ram it through there somehow. The stock comfy cloud (where BCE runs) won’t run it, so no love for Mac folks short term. Hopefully that can be fixed 
Yeah. I think if you were to use proven good and fast Sammie based models in various applications and than use LTX for those shots where you have semi transparency that would be best combo. After all LTX has pretty hefty price and limitations are pretty server compared to established other models so its an overkill to use it for everything at this point anyway.
Did some LTX 2.5 on my local machine yesterday. Getting similar results and will echo that both LTX 2.5 and Sammie have pros and cons, but both are worthy of inclusion in our bag of tools.
How are you using Sammie? In comfyUI or the standalone app?
For LTX Alpha Gen I’ve just been testing it, I happen to be on a green screen job, girl with blowing hair etc. It’s amazing! Better than a key I think. Have we reached a point where we just tell the client to shoot against a screen similar in colour to the eventual BG, or a grey screen. Seems like spill is more of an issue now than separating FG elements from BG?
I think that is one of the pitches by LTX, that it could be a key replacement.
One thing that LTX seems to excel at is keeping the mask through changing conditions. Neither flares or object going from sun to shade in the examples above trip it up.
But you have less control on what is fg vs bg. So you may need to pair it with something to help guide that or do matte ops. Onyx has quite the number of controls, which may be a good pairing for those complex scenes and less about transparency.
I think everyone will find their combo that works in their pipeline / hardware.
The one that does keep coming up short is AutoMatte. Convenient, but rather crude at this point without temporal consistency, limited fg/bg guidance, and frequent rough edges or holes. Good enough for quick garbage mattes, or super simple shapes in ideal conditions.