FOR THE PEOPLE ASKING: Where do I add the LORAS? Here is a summary:
Bold Summary: Add the official LTX 2 Detailer LoRA (IC-LoRA) primarily in the Second Pass (Refinement/Upscale) to enhance texture without altering composition. Add Third-Party LoRAs (like "Better Body") to BOTH passes to ensure the subject's structure is generated correctly and maintained during refinement.
- The LTX 2 Detailer LoRA (Official IC-LoRA)
Recommendation: Second Pass (or Both)
This LoRA (ltx-2-19b-ic-lora-detailer.safetensors) is designed to add fine texture and "crispness" to the footage.
First Pass: You can add it here, but it is less critical. The first pass is about establishing motion and composition.
Second Pass (Refinement): This is the most important place for it. As the model upscales or refines the video (e.g., in a "Hi-Res Fix" stage), the Detailer LoRA forces the model to resolve fine textures (skin, fabric) that often get smoothed out.
Strength: Keep it around 0.5 – 0.8. Setting it too high (1.0+) in the second pass can sometimes cause "shimmering" or over-sharpened artifacts.
- Third-Party LoRAs (e.g., "Better Body", Styles, Characters)
Recommendation: BOTH Passes
These LoRAs define the structure, shape, and identity of your subject.
First Pass: Mandatory. You must have the LoRA active here so the model knows what to generate (e.g., the specific body shape) from the very beginning noise. If you omit it here, the base composition will be wrong.
Second Pass: Mandatory. You must keep it active here (usually at the same or slightly lower strength) so the model doesn't "hallucinate" the subject away or revert to the generic LTX body type during the denoising process.
How to Connect Them (ComfyUI)
In a standard two-pass workflow, you typically have a chain of Model Loaders.
Chain the LoRA Loaders: Connect LoraLoaderModelOnly nodes in a sequence after your checkpoint loader.
Checkpoint Loader -> Body LoRA -> Detailer LoRA -> Model Output.
Apply to Samplers:
Feed the final model output (with all LoRAs attached) into Sampler 1 (First Pass).
Feed the same model output into Sampler 2 (Second Pass).
Advanced: If you strictly want the Detailer only in the second pass, you would need to branch the model link:
Branch A (Body LoRA only) -> Sampler 1.
Branch B (Body LoRA + Detailer LoRA) -> Sampler 2.
Simplest Method: Just chain them all and feed the result to both samplers. It works perfectly fine for 95% of cases.
It would be great if you can share a reference workflow, to see all the connections in places.
Bold Summary: Add the official LTX 2 Detailer LoRA (IC-LoRA) primarily in the Second Pass (Refinement/Upscale) to enhance texture without altering composition. Add Third-Party LoRAs (like "Better Body") to BOTH passes to ensure the subject's structure is generated correctly and maintained during refinement.
- The LTX 2 Detailer LoRA (Official IC-LoRA)
Recommendation: Second Pass (or Both)
This LoRA (ltx-2-19b-ic-lora-detailer.safetensors) is designed to add fine texture and "crispness" to the footage.First Pass: You can add it here, but it is less critical. The first pass is about establishing motion and composition.
Second Pass (Refinement): This is the most important place for it. As the model upscales or refines the video (e.g., in a "Hi-Res Fix" stage), the Detailer LoRA forces the model to resolve fine textures (skin, fabric) that often get smoothed out.
Strength: Keep it around 0.5 – 0.8. Setting it too high (1.0+) in the second pass can sometimes cause "shimmering" or over-sharpened artifacts.
- Third-Party LoRAs (e.g., "Better Body", Styles, Characters)
Recommendation: BOTH PassesThese LoRAs define the structure, shape, and identity of your subject.
First Pass: Mandatory. You must have the LoRA active here so the model knows what to generate (e.g., the specific body shape) from the very beginning noise. If you omit it here, the base composition will be wrong.Second Pass: Mandatory. You must keep it active here (usually at the same or slightly lower strength) so the model doesn't "hallucinate" the subject away or revert to the generic LTX body type during the denoising process.
How to Connect Them (ComfyUI)
In a standard two-pass workflow, you typically have a chain of Model Loaders.
Chain the LoRA Loaders: Connect LoraLoaderModelOnly nodes in a sequence after your checkpoint loader.Checkpoint Loader -> Body LoRA -> Detailer LoRA -> Model Output.
Apply to Samplers:
Feed the final model output (with all LoRAs attached) into Sampler 1 (First Pass).
Feed the same model output into Sampler 2 (Second Pass).Advanced: If you strictly want the Detailer only in the second pass, you would need to branch the model link:
Branch A (Body LoRA only) -> Sampler 1.
Branch B (Body LoRA + Detailer LoRA) -> Sampler 2.
Simplest Method: Just chain them all and feed the result to both samplers. It works perfectly fine for 95% of cases.