🚀 MrXin LTX 2.3 T2V EROS V1
This is the brand new Text-to-Video version of the MrXin LTX 2.3 EROS workflow.
Built from the ground up for pure text-to-video generation, V1 delivers the same high-motion quality, perfect audio sync and low-VRAM performance you know from the I2V versions — but now without needing a starting image. Just drop your prompt and generate cinematic, explicit videos.
V1 Key Features
Full text-to-video pipeline with strong, natural motion from the first frame
Separate Manual Sigmas for First Pass and Final Pass
Easy model switching between original Eros checkpoint and Distilled 22B model
Powerful default LoRA stack with all your favorite NSFW/animation triggers pre-loaded
Built-in Video Editor: RTX Super Resolution + nmkdSiaxCX model upscaler + RIFE interpolation to 48 FPS
Two live previews + clean output folders
Excellent results on 12 GB cards with 32 GB RAM
✅ What You Get
Strong, fluid motion driven purely by your text prompt
Smooth 24 FPS MP4 with perfectly synced audio (moans, impacts, wet sounds)
High-quality explicit animation with accurate anatomy and physics
Optional 48 FPS final output via built-in editor
Easy toggle to skip Final Pass for faster test renders
📥 Required Models + Direct Download Links (Click name to download)
Checkpoint → ltx2310eros_beta.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-x2-1.1.safetensors
Model Upscaler → nmkdSiaxCX_200k.safetensors
Distilled LoRA → ltx-2.3-22b-distilled-lora-dynamic_fro09_avg_rank_105_bf16.safetensors
📁 Folder Structure (same as I2V versions)
text
ComfyUI/
├── models/
│ ├── checkpoints/ ← ltx2310eros_beta.safetensors
│ ├── diffusion_models/ ← 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-dynamic_fro09_avg_rank_105_bf16.safetensorsPro Tips
Use detailed, cinematic prompts — the workflow loves long, descriptive text
Faster tests? Disable the Final Pass group and save only the First Pass
Cloud users? Use the Model Upscaler (bypass RTX node)
This T2V version finally brings the full EROS experience to pure text prompts.
Ready to generate! 🔥
— MrXin (April 2026)
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.
Description
🚀 MrXin LTX 2.3 T2V EROS V2
This is a fully-featured, production-ready Text-to-Video workflow for LTX 2.3 (Eros) that generates high-quality 20–25 second cinematic videos from text prompts only — no starting image required.
V2 brings improved motion consistency, better prompt adherence, and refined post-processing compared to V1.
🔥 Key Features
Pure Text-to-Video generation with strong, natural motion
Dual model support: 10Eros V1 FP8 + Distilled 22B model with easy switching
Separate Manual Sigmas for First Pass and Final Pass (excellent motion preservation)
Heavy NSFW/animation optimized LoRA stacking (Power Lora Loader)
Built-in Video Editor section with RTX Super Resolution, nmkdSiaxCX upscaler & RIFE interpolation to 48 FPS
Baked audio with synchronized moans, impacts and breathing
Two live previews + clean output folders
Optimized for RTX 5080 / 5090 and lower VRAM setups
✅ What You Get
Smooth 24 FPS MP4 with synced audio
Strong cinematic motion that survives the Final Pass
High detail, realistic physics and excellent anatomy
Easy toggles to disable Final Pass or Video Editor for faster generation
📥 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
Pro Tips
Use detailed, cinematic prompts with action descriptions and sound cues
For maximum motion: lower Final Pass strength to 0.6–0.7
Disable Final Pass + Editor for quick tests
Best results with the new 10Eros V1 FP8 model — highly recommended
Just drop your prompt, adjust sliders, and generate stunning text-to-video clips.
Ready for CivitAI! 🔥
— MrXin (May 2026)
FAQ
Comments (10)
Not to say Dr3aml4y isn't good, but it is certainly not a default lora needed especially since it's just a generalized lora thats being stacked on a generalized model and it is bringing base model influence with it. You'd want penile praxis + maybe synthpussy or nudity/anatomy loras for a stack instead.
Getting the error
AttributeError: 'Gemma3TextConfig' object has no attribute 'rope_local_base_freq'
The final video always ends up with some crazy like tiled pattern background and distorts the main image. Like the first pass is good, the second/final pass ruins the vid.
Do I have to use kj nodes or can i just use the normal ones?
if anyone is on Blackwell V1 will work V2 will not (well it didn't for me) My setup RTX 5080 16gb VRAM 32gb RAM , im using blackwell master build fully updated 2.13.0.dev20260519+cu130 NVIDIA GeForce RTX 5080 13.0. but the difference is :
V1 uses VAEDecodeTiled with settings [512, 64, 2048, 8] — a standard tiled VAE decoder.
V2 replaced it with LTXVSpatioTemporalTiledVAEDecode — the custom LTX spatio-temporal decoder which is a kjnodes node.
LTXVSpatioTemporalTiledVAEDecode is broken on my setup. VAEDecodeTiled works fine.
The fix is inside the subgraph — replace LTXVSpatioTemporalTiledVAEDecode (node 149) with a standard VAEDecodeTiled node using V1's settings:
tile_size = 512
overlap = 64
temporal_size = 2048
temporal_overlap = 8
PLEASE HELP ME !!
LTXVAudioVAEEncode
Input type (float) and bias type (struct c10::BFloat16) should be the same
is there a way to change step number
how can I change to video to landscape?
Hello. Thanks for the wf. by the way, i've a very bluring video. Is there a chance to restore a clean video ? i've an RTX 4070 super
What i can say the t2v is working pretty well ;)