Got an error; the output is just visual noise. Why?
root=/xxx/xx/
python -m ltx_pipelines.ti2vid_two_stages \
--checkpoint-path ${root}/LTX-2.3/ltx-2.3-22b-dev.safetensors \
--distilled-lora ${root}/LTX-2.3/ltx-2.3-22b-distilled-lora-384.safetensors 0.8 \
--spatial-upsampler-path ${root}/LTX-2.3/ltx-2.3-spatial-upscaler-x2-1.0.safetensors \
--gemma-root ${root}/gemma-3-12b-it-qat-q4_0-unquantized \
--prompt "A beautiful sunset over the ocean" \
--output-path output_v1.mp4
python -m ltx_pipelines.ti2vid_one_stage \
--checkpoint-path ${root}/LTX-2.3/ltx-2.3-22b-dev.safetensors \
--gemma-root ${root}/gemma-3-12b-it-qat-q4_0-unquantized \
--prompt "A beautiful sunset over the ocean" \
--output-path output_v2.mp4
output_v1.mp4:
output_v2.mp4:
1: The 1.0 Upscaler is broken - use 1.1
2: The QAT text encoder from Sikaworld is broken. Use HereticX instead.
3. Provide us with the workflow that you are using.
1: The 1.0 Upscaler is broken - use 1.1
2: The QAT text encoder from Sikaworld is broken. Use HereticX instead.
3. Provide us with the workflow that you are using.
Load HereticX weights from the link below？
https://huggingface.co/Sikaworld1990/gemma3-12B-hereticx-sikaworld-ltx-2/tree/main
Yes, that's the one. I also found this model works better if you have say a 16GB card: ltx-2.3-22b-dev_transformer_only_fp8_scaled. If you have a 5090, then the regular dev should be OK to use.
FileNotFoundError: No files matching pattern 'tokenizer.model' found under Sikaworld/gemma3-12B-hereticx-sikaworld-ltx-2 ?
Is this model suitable for video generation tasks compatible with LTX2.3, or is it limited to LTX2 tasks?
It works fine with the LTX 2.3. You have to edit the file paths in your workflow to point it to the right locations for your files.
The proper file is gemma3-12B-Hereticx-sikaworld.safetensors - it goes in your text_encoder directory.