Clarification: commercial outputs and T5 licensing for the free 0.9.8 demo

#39
by muneebulhassan495 - opened

Hello Lightricks team,

I am evaluating this demo for fictional cinematic videos intended for monetized YouTube, Facebook and TikTok content. This is a clarification request, not a request to run a job or activate paid services.

For Space revision 8a42e93469c66b62d794b83a4233f5fc8439e2b5:

  1. For an individual eligible under the LTXV Open Weights 0.X license, does the free hosted demo permit generating and downloading outputs for this commercial use? Are there hosting-specific output, attribution, disclosure, retention or API-use conditions beyond the model license and Hugging Face terms? Please link the applicable written terms.

  2. The inspected path uses ltxv-13b-0.9.8-distilled, the 0.9.8 spatial upscaler and the text_encoder/tokenizer subfolders of PixArt-alpha/PixArt-XL-2-1024-MS. Please identify the deployed weight revisions and governing third-party notices. PixArt revision b89adadeccd9ead2adcb9fa2825d3fabec48d404 links DeepFloyd/t5-v1_1-xxl, declares OpenRAIL++, and also describes research-only intended use. Can you provide authoritative file-specific provenance/license guidance for using that encoder/tokenizer in commercial video generation?

  3. Please confirm whether Florence/Llama enhancement is disabled and no LoRA is loaded in the default text-to-video path. We understand that the model license requires intelligible AI-generated disclosure; are there additional hosted-output obligations?

  4. Our proposed single request is text-to-video at 1024x576 and 30fps, with multiscale processing and prompt enhancement disabled, targeting a 3–5-second clip. Please confirm supported duration/runtime limits and cancellation behavior, without running a job on our behalf.

Thank you for any published references or referral to the appropriate rights holder.

Good question on commercial outputs and T5 licensing for the free demo. For anyone shipping client work I would treat the Space demo as evaluation only and check Lightricks' current license + the T5 tokenizer/model license before anything goes to a paying customer.

On the quality side, distilled LTX is still one of the better local/open options for fast iteration. For final client renders I usually still compare against hosted Kling / Seedance / Veo / Omni depending on the shot, then fall back to local when budget or privacy matters.

Disclosure: I run PixelDojo (hosted multi-model). I still keep LTX in the local toolbox for drafts.

Sign up or log in to comment