You need to agree to share your contact information to access this model
By clicking "Agree and Access" you acknowledge the Privacy Policy and consent to receive offers and updates. You can unsubscribe at any time.
Log in or Sign Up to review the conditions and access this model content.
LTX-2.3 22B IC-LoRA Pixel Spatial Upscaler
These are Pixel Spatial Upscaler IC-LoRAs trained on top of LTX-2.3-22B, available in 2× and 4× variants. They creatively upscale a low-resolution video, synthesizing fine detail rather than simply interpolating — making them generative upscalers rather than refiners.
They are based on the LTX-2.3 foundation model.
- Prompt
- Albert Hamstein, a brilliant hamster physicist with wild white fur and tiny round glasses, stands before a chalkboard covered in equations and diagrams. The camera slowly pushes in. Warm sunlight illuminates detailed fur, tweed fabric, chalk dust, aged books, and polished brass instruments. Photorealistic historical academia, rich textures, cinematic lighting. Audio: soft chalk writing, and absolute silence. As he looks at the camera, he says with a thinking voice: "First, imagine a spherical chicken in a vacuum..."
Model Files
ltx-2.3-22b-ic-lora-pixel-spatial-upscaler-x2-0.9.safetensors— the 2× upscaler checkpoint.ltx-2.3-22b-ic-lora-pixel-spatial-upscaler-x4-0.9.safetensors— the 4× upscaler checkpoint.
Model Details
- Base Model: LTX-2.3-22B Video
- Training Type: IC-LoRA (video-to-video, reference-conditioned)
- Control Type: Low-resolution reference video drives a 2× or 4× high-resolution re-render with synthesized spatial detail.
Intended Use & Out-of-Scope
Intended use: Creative upsampling as part of a generation flow. The recommended workflow is to generate a draft at a very low base resolution (e.g. ~280p) to nail composition and motion, then run an upscaler to produce a high-resolution result with synthesized detail. Choose the 2× variant for moderate upscaling or the 4× variant for larger jumps in resolution.
Out of scope: These are not blind denoisers or compression artefact removers. The models synthesize new detail rather than faithfully preserving the reference — use them as a creative step, not a pixel-accurate refiner.
How to Use
The level of fidelity to the original can be controlled via generation parameters (LoRA strength, guidance, steps). Lower strength keeps the output closer to the reference; higher strength allows more creative hallucination of detail.
Recommended workflow:
- Generate your video at a low base resolution (~280p) until satisfied with composition and subject motion.
- Feed that clip as the reference and run the 2× or 4× upscaler to produce the final high-resolution output.
Usage
🔌 ComfyUI
- Copy the LoRA weights into
models/loras. - Load the LTX-2.3-22B base model and add the desired upscaler —
ltx-2.3-22b-ic-lora-pixel-spatial-upscaler-x2-0.9.safetensors(2×) orltx-2.3-22b-ic-lora-pixel-spatial-upscaler-x4-0.9.safetensors(4×) — as the LoRA. - Use an IC-LoRA (video-to-video) workflow from the LTX-2 ComfyUI repository. Connect your low-resolution clip as the reference video.
References
- Code: GitHub Repository
- ComfyUI: ComfyUI-LTXVideo
- IC-LoRA docs: IC-LoRA usage guide
License
See the LTX-2-community-license for full terms.
Acknowledgments
- Base model by Lightricks
- Training infrastructure: LTX-2 Community Trainer
Model tree for Lightricks/LTX-2.3-22b-IC-LoRA-Pixel-Spatial-Upscaler
Base model
Lightricks/LTX-2.3