🚀 MrXin LTX 2.3 I2V EROS V6
This workflow has continuously evolved based on user feedback to deliver one of the strongest LTX 2.3 I2V experiences available. Thx everyone for the feedbacks.
V6 is a major leap forward, focusing on stronger motion preservation, better prompt adherence, and more user control while keeping excellent stability and VRAM efficiency.
🔥 What’s New in V6 (Compared to V5)
Added LTX2 NAG (Negative Attention Guidance) on both First Pass and Final Pass for significantly improved prompt following and motion quality.
Switched to VAELoader KJ nodes for Video VAE and Audio VAE (better compatibility and stability).
Improved Final Pass by repositioning the LTXVCropGuides node for cleaner latent processing.
Added flexible sigma system: Manual Sigmas (8 steps) or automatically generated sigmas via LTXVScheduler + Sigmas Sigmoid (12 steps).
Final Pass is Manual Sigmas only for stability.
Audio volume default set to 0 (neutral) with easy slider control.
Enhanced image conditioning using LTXVImgToVideoInplaceKJ in both passes.
Cleaner workflow layout, better group organization, and more helpful in-workflow tips.
✅ Core Features
Extremely strong, natural and consistent motion from First to Final Pass
Excellent anatomy, lighting, physics and NSFW detail
High-quality 24 FPS video with perfectly synced audio
Easy model switching between 10Eros V1 FP8 and Distilled 22B
Built-in Video Editor (RTX Super Resolution + nmkdSiaxCX upscaler + RIFE 48/60 FPS)
Optimized for 16GB VRAM setups and 32GB RAM
📥 Required Models + Direct Download Links
10Eros V1 FP8 (Main Checkpoint) → ltx2310eros_v1_FP8.safetensors
Distilled Model → ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled_v3.safetensors
Text Encoder → gemma_3_12B_it_fp8_e4m3fn.safetensors
Text Projection → ltx-2.3_text_projection_bf16.safetensors
Video VAE → LTX23_video_vae_bf16.safetensors
Audio VAE → LTX23_audio_vae_bf16.safetensors
Preview VAE → taeltx2_3.safetensors
Spatial Upscaler → ltx-2.3-spatial-upscaler-x2-1.0.safetensors
Model Upscaler → nmkdSiaxCX_200k.safetensors
Distilled LoRAs (First & Second Pass)
📁 Folder Structure
ComfyUI/
├───📂 models/
│ ├───📂 diffusion_models/
│ │ └─── ltx2310eros_beta.safetensors
│ │ └─── ltx-2.3-22b-distilled_transformer_only_fp8_input_scaled_v3.safetensors
│ │
│ ├───📂 text_encoders/
│ │ └─── gemma_3_12B_it_fp8_e4m3fn.safetensors
│ │
│ ├───📂 clip/
│ │ └─── ltx-2.3_text_projection_bf16.safetensors
│ │
│ ├───📂 VAE/
│ │ └─── LTX23_video_vae_bf16.safetensors
│ │ └─── LTX23_audio_vae_bf16.safetensors
│ │ └─── taeltx2_3.safetensors
│ │
│ ├───📂 latent_upscale_models/
│ │ └─── ltx-2.3-spatial-upscaler-x2-1.1.safetensors
│ │
│ ├───📂 upscale_models/
│ │ └─── nmkdSiaxCX_200k.safetensors
│ │
│ └───📂 Lora/
│ └─── ltx-2.3-22b-distilled-lora-1.1_fro90_ceil72_condsafe.safetensors
│ └─── ltx-2.3-22b-distilled-lora-384-1.1.safetensors
Custom Nodes Overview – MrXin LTX 2.3 I2V EROS V6
ComfyUI-KJNodes
rgthree-comfy
ComfyUI-easy-use
ComfyUI-mxToolkit
ComfyUI-VideoHelperSuite
ComfyUI-LTXVideo
controlaltai-nodes
comfyui_nvidia_rtx_nodes
GACLove/ComfyUI-VFI (voor RIFE)
comfyui_memory_cleanup + comfyui-impact-pack
Pro Tip: Use the new 10Eros V1 FP8 model with its recommended trigger words for the best results.
✅ Quick OOM Fix Guide
If you’re getting Out of Memory errors or weird faces in the final pass, follow these steps in order:
1. Update or Fresh Install ComfyUI + Install Missing Custom Nodes
Download and run the Easy Installer: 👉 https://github.com/Tavris1/ComfyUI-Easy-Install
Or watch this short video: 👉 https://www.youtube.com/watch?v=CgLL5aoEX-s&t=1041s
After updating/installing:
Open ComfyUI → go to Manager → click Install Missing Custom Nodes
Restart ComfyUI completely.
2. Add the Low VRAM Flags
Go to your ComfyUI folder.
Find the file ComfyUI.bat.
Right-click → Edit (open with Notepad).
Find the line that has ComfyUI\main.py .
Add the flags at the end so it looks like this:
bat
ComfyUI\main.py --lowvram --reserve-vram 6 --preview-method none --disable-xformers --disable-smart-memorySave and launch ComfyUI using this .bat file.
3. Increase Windows Swap File (Virtual Memory) This helps when VRAM still spikes:
Right-click This PC → Properties → Advanced system settings.
Click Settings under Performance → Advanced tab → Change (under Virtual memory).
Uncheck “Automatically manage paging file size for all drives”.
Select your main drive (usually C:) → choose Custom size.
Set:
Initial size (MB): 49152
Maximum size (MB): 98304
Click Set → OK → restart your PC.
After doing all three steps, reload the workflow and test with Longer Side 1024 and Video Length 20 seconds first. This combination fixes OOM for almost everyone on 12GB cards.
Version History & Previous Updates
V5 — Major base model upgrade to 10Eros V1 FP8. Significantly better realism, anatomy, lighting and motion coherence. Updated model loading and separate LoRA handling for each pass.
V4 — Introduced separate Manual Sigmas for First and Final Pass, much stronger Final Pass motion, full TwoWaySwitch system for model selection, expanded LoRA stack, and refined Video Editor.
V3 — Solved the biggest V2 complaint by preserving much more motion in the Final Pass. Added Model Upscaler (nmkdSiaxCX), option to disable Final Pass, distilled model support, and custom resolution.
V2 — Added full built-in Video Editor (RTX Super Resolution + RIFE to 48 FPS), improved LoRA suite, better audio sync, and major stability enhancements.
V1 — The original production-ready workflow. Dual-pass system, strong audio integration, excellent low VRAM optimization (12GB cards), and live previews.
Disclaimer:
This workflow is provided for entertainment, artistic, and creative purposes only. It may not be used for any illegal, harmful, non-consensual, or malicious activities. Please use it responsibly and respect all applicable laws and ethical guidelines.
— MrXin (May 2026) 🔥
Description
🚀 MrXin LTX 2.3 I2V EROS V6 - Version Changes
Major Updates in V6:
Added LTX2 NAG (Negative Attention Guidance) nodes for both First Pass and Final Pass – greatly improves prompt adherence, motion quality, and overall coherence.
Switched to VAELoader KJ nodes for loading both Video VAE and Audio VAE for better compatibility and stability.
Improved Final Pass significantly by repositioning the LTXVCropGuides node for cleaner latent processing.
Added flexible sigma system for first pass:
Manual Sigmas (8steps)
Or automatically generated sigmas using LTXVScheduler + Sigmas Sigmoid with 12 steps (default). you can adjust these steps.
Final Pass is now locked to Manual Sigmas only for maximum stability and motion preservation.
Audio volume default changed to 0 (neutral) with easy manual adjustment available.
Enhanced image conditioning using LTXVImgToVideoInplaceKJ in both passes with improved strength control.
This version focuses on stronger motion retention from First to Final Pass, much better prompt following thanks to NAG, and more user control over sigmas.
FAQ
Comments (45)
Any tips/guide on prompt structure for a clean output?
i use grok for my prompts, i upload a image and then this text as example "Can you create an NSFW play script for me in English describing sex action and sounds? It is for a 20-second scene. A woman is having sex with a man lying beneath her, and the woman is giving a blowjob to a man standing in front of her. The woman has an orgasm. Add an outdoor background sound over the entire scene. Make the script unique. The man says dirty things to the woman. Make the dialogue a maximum of 4 sentences." Change the action in to what you want.
Been trting to get this workload to work since V3 and just cant..output is bad doesnt adhere to prompt and gets artifacts...im on rtx5080 using blackwell master comfy ui i disabled sage as per tip in v6.. i follow the tips on each workflow , dont change any settings apart from what it says i use copy ,/paste prompts from examples here the only difference would be img since i dont have your input img. BUT results are horrible not even remotely close to what you guys present with each release...what am i doing wrong?
I recommend installing the complete version of ComfyUI. I’ve tested it many times and consistently achieved very good results. Use the link provided in the workflow to install it. It includes the latest ComfyUI updates, as well as additional add-ons such as SageAttention.
@MrXin blackwell is optimised for rtx50 standard causes loads of issues. BUT i just got v3 first pass working it produced almost good video which tells me its something else. I run json through gemini and its suggesting some encoding issue. The workflow is heavily leaning towards 3060 and some settings might be too aggresive. Also 2nd pass always comes out noisy, blurry. Could this be encoding and decoding issue? Is there a way you can adjust it or tell me what to adjust for blackwell/5080 please?
I dont mind testing if you dont have access to one
Error with v6. Hopefully somebody can help me. I’ve been able to use every workflow up until v5 with no issue. But with V6, I am getting a VAE loader KJ error. It gives a message that takes two positional arguments, but three were given. I’m not sure what that means. Thanks for all your hard work. Love the workflows.
try to reload the node be right clicking on it. If that doesn't help, update comfyui.
@MrXin Thanks I’ll give that a tray 😊 love your work. I’ve used almost every workflow that has been uploaded and I by far prefer yours. It gives me the best results. I really appreciate you taking the time to make these and share them.
@nutmegzero439 Thx, man. If you still have a problem with the KJ VAE node, change it with a regular VAE node from the V5 workflow.
@MrXin Thank you I was able to get it working by updating comfy UI. However, now when I generate on v5 or v6 video starts clear, but then goes really blurry by the end. But if I disable chunk it fixes it. Any workaround for that or am I OK not using chunk I’m not even sure what that is lol I’m kind of a noob
V6 seems to be much slower than V5. Is that expected?
Does exactly what was on the box: better prompt adherence and more movement. Didn't notice any drop in clarity, maybe a little slower, but still perfectly fine. V6 is another banger. I was just about to try some of the new face consistency nodes on V5, i'll try them out here. Or maybe something we'll see in V7?
Upload the finished video and get an error in the upscale attempt. The error is that it's impossible to proceed due to astronomical memory requirements. Does anyone know the cause or know a solution?
I'm getting: AttributeError: 'function' object has no attribute 'to'
On the LTX2 NAG First Pass node.
Fixed by manually updating nodes. ComfyUI's auto updates didn't get the right versions I guess...
Hello!
First of all thank you for your work, for clear and conscious instructions.
Can you please clarify one thing though - there are 2 loras listed as required in the instructions but i see only one (fro90) being used in the workflow. Am i missing something?
If you enlarge the “Load Distilled Lora Final Pass” node, you’ll see that I use ltx-2.3-22b-distilled-lora-384-1.1.safetensors for the final pass. This helps prevent weird subtitles or burn-in artifacts at the end of the video.
hey would be nice if you can add ltx director and face fix (likeness anchor) by eros also middle frame end frame support @MrXin thanks your workflow are amazing
These are the two things I have been trying and failing to implement myself, I'm useless at making workflows, LTX director is a game changer so I hope we see it coming to this brilliant wf!
why can't we put 720 as a height/width value?
Great workflow but curious about this.
Hey, this is because LTXV 2.3 can only process resolutions in steps of 32.
720 / 32 = 22.5, which is not a whole number, so this resolution is not compatible with LTXV 2.3.
Cool makes sense and thanks again for the workflow. works great :)
Why am I not seeing any thumbnails when generating videos? I don't have this issue with other workflows.
thx for the info, i will remove them from the workflow.
537 sec here to get to 1st pass preview on a 5090. gets stuck at 3% for a while. V5 was way faster 99 seconds
Hey, I’m really sorry about that. The dual clip loader device was set to "CPU" and it needs to be "default". I was testing some settings and forgot to change it back. I’ll make a V6.1 update.
V6.1 is online.
Went back to V6 and changed to "default" now 90 seconds V6.1 is down to 75sec. Great workflow thx.
for me it take something like 20-40 min but i think cz i use loras in the wrong way i want use more loras but the more the worst quality the longer it take i think
Amazing workflow! But with the v6 while the quality of the video and prompt coherence have been improved, the gen time also sadly has gone up quite a bit. I also hit a CPU max load and a GPU out of memory error, despite scaling down the video, so i think it needs a little adjustment
I’m not sure if this is correct or not, but I was having somewhat of a similar issue. I could just hear my fans screaming. I noticed on V6 under dual clip loader device was set to CPU on v5 it’s set to default. So I changed it to default on V6 and it seemed to fix that problem with no noticeable changes. Again I am a noob so I don’t know if that is a good setting to change, but it worked for me.
@nutmegzero439 Hey, yeah, this should be set as default. I was testing some settings and forgot to change it back. I’m really sorry about that. I’ll make a V6.1 update.
@MrXin Works fine now, ty for quick update!
@MrXin Did a little more testing and it seems that the performance is still very heavy. I noticed that on the v6 version of the workflow my PSU fans start to kick in to compensate for the load on my GPU and CPU, which didn't happen with the v5 version. Seems to happen more with Final Video enabled. This happenes even with the CPU fix :/
@biggchungus1337 Well, I moved the “LTXVCropGuides” node in the final pass to improve the quality. That might be causing this issue. However, my generation times are still the same as in the previous versions.
What is your recommendings for 3090 rtx 48 gb ram. Also my characters face changing how can i preserve the face consistency ?
Add a prompt that says something like "Preserving the original face in the picture".This works for me.
With that system you dont need to worry much about settings if you use the fp8 version.
anyone else had this mysterious issue with the final stage, in the first stage the video looks fine, but the final stage becomes a tiled washed out mess that looks totally different. Maybe it's just me
Update or reinstall ComfyUI using ComfyUI EZI. Use the links on the left side of the workflow for the download link and the explanation video.
I did i wil try reinstalling comfy UI
@polfa757 lower the strength on the "Load Distilled Lora Final Pass". The default value of 0.5 seems too high. I changed it to 0.3, and it looks much better now.
@craempei i will try that, are you on blackwell as well? I tried so many workflows from mrxin and anything past v3 just doent work v3 is the only one that goves me clean nice 1st pass. Almost perfect but when i add final it breaks so i was trying newer workflows including t2v (on that one v1 eorks and v2 doesnt)
Lol I just took another look today and I noticed that I had chosen the ltx 2 spatial upscaler instead ot the ltx 2.3 one lol, so all good now, thanks for the workflow.