Nodes/h3-relay/H3 Relay · LTX Upscale Model Loader
ComfyUI Node

H3 Relay · LTX Upscale Model Loader

A ComfyUI node in H3 Relay/loaders with 10 inputs and 1 output.

By akatz-ai·Created 8 days ago·Updated 4 days ago· 9
H3 Relay · LTX Upscale Model Loader
    • ltx_model
    model_nameltx-2.5-22b-dev-transformer-comfy-int8-convrot.safetensors
    vae_nameltx-2.5-video-vae-bf16.safetensors
    latent_2x_model_nameltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors
    text_encoder_namegemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot.safetensors
    distilled_loraltx-2.5-22b-distilled-lora-450-bf16.safetensors
    distilled_strength1.00
    pixel_upscale_ic_loraltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensors
    pixel_upscale_ic_strength1.00
    weight_dtypedefault
    manual_cache_revisionv1
    CategoryH3 Relay/loaders

    Inputs (10)

    NameTypeDefaultDescription
    model_nameCOMBOltx-2.5-22b-dev-transformer-comfy-int8-convrot.safetensors0 options:
    vae_nameCOMBOltx-2.5-video-vae-bf16.safetensors0 options:
    latent_2x_model_nameCOMBOltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensorsLearned 2x latent spatial expansion. This establishes the high-resolution target latent before diffusion refinement.
    text_encoder_nameCOMBOgemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot.safetensors0 options:
    distilled_loraCOMBOltx-2.5-22b-distilled-lora-450-bf16.safetensorsAdapts the LTX dev transformer for the fast low-step schedule. It does not perform the spatial upscale.
    distilled_strengthFLOAT1.00-4–4
    pixel_upscale_ic_loraCOMBOltx-2.5-22b-ic-lora-pixel-spatial-upscaler-x2-1.0.safetensorsReference-conditioned IC-LoRA. It guides diffusion from the original low-resolution pixel video after the latent has been expanded 2x.
    pixel_upscale_ic_strengthFLOAT1.000–2
    weight_dtypeCOMBOdefault4 options: default, fp8_e4m3fn, fp8_e4m3fn_fast, fp8_e5m2
    manual_cache_revisionSTRINGv1Manual cache reset only. Change this when replacing model contents without changing the filename; normal model, LoRA, dtype, strength, and attention changes are tracked automatically.

    Outputs (1)

    NameTypeDescription
    ltx_modelH3_RELAY_MODEL