Summary
Requesting an MLX port / pipeline for legacy LTX-Video (0.9.x) (e.g. Lightricks/LTX-Video, Lightricks/LTX-Video-0.9.x-dev) — the pre-LTX-2 model family — so it can run natively on Apple Silicon alongside the existing ltx_2 pipeline.
Why
mlx-video currently ships only ltx_2 (the newer LTX-2 / LTX-2.3 family) and wan_2 (confirmed under mlx_video/models/). The older LTX-Video 0.9.x checkpoints are still widely used and downloaded, but there is no native MLX path for them. Note: this is distinct from the LTX-2 family — issue #37 etc. cover LTX-2.3 loading; this request is specifically for the 0.9.x lineage.
Without a port, downstream Mac inference servers (we run fusion-mlx on top of mlx-video) cannot serve legacy LTX-Video checkpoints and must decline the request.
What's needed
An MLX implementation of the legacy LTX-Video 0.9.x pipeline:
- VAE (encoder/decoder) in MLX
- DiT transformer in MLX
- The LTX-Video 0.9.x scheduler in MLX
- A
generate_video(...) entrypoint mirroring the ltx_2/wan_2 signatures
Context
In our downstream server we currently expose legacy LTX-Video as an explicitly unimplemented backend that auto-detects ltx-video/ltx_video model paths and fails loudly pointing here. We would love to replace the NotImplementedError with a real delegator to mlx-video once a port lands. (Users wanting the newer model are already directed to the working ltx_2 backend.)
Happy to help test a WIP port on M-series hardware if anyone picks this up.
Summary
Requesting an MLX port / pipeline for legacy LTX-Video (0.9.x) (e.g.
Lightricks/LTX-Video,Lightricks/LTX-Video-0.9.x-dev) — the pre-LTX-2 model family — so it can run natively on Apple Silicon alongside the existingltx_2pipeline.Why
mlx-video currently ships only
ltx_2(the newer LTX-2 / LTX-2.3 family) andwan_2(confirmed undermlx_video/models/). The older LTX-Video 0.9.x checkpoints are still widely used and downloaded, but there is no native MLX path for them. Note: this is distinct from the LTX-2 family — issue #37 etc. cover LTX-2.3 loading; this request is specifically for the 0.9.x lineage.Without a port, downstream Mac inference servers (we run fusion-mlx on top of mlx-video) cannot serve legacy LTX-Video checkpoints and must decline the request.
What's needed
An MLX implementation of the legacy LTX-Video 0.9.x pipeline:
generate_video(...)entrypoint mirroring theltx_2/wan_2signaturesContext
In our downstream server we currently expose legacy LTX-Video as an explicitly unimplemented backend that auto-detects
ltx-video/ltx_videomodel paths and fails loudly pointing here. We would love to replace theNotImplementedErrorwith a real delegator to mlx-video once a port lands. (Users wanting the newer model are already directed to the workingltx_2backend.)Happy to help test a WIP port on M-series hardware if anyone picks this up.