Loop Forge

Heavily macroblocked market

Works, from LTX 2.5 Creative Lab LoRAs, tested locally

Frame from the Blocky input (120 kbps) result
Blocky input (120 kbps)
Frame from the Original result
Original
Frame from the Decompression LoRA result
Decompression LoRA
Watch it in the video at 9:09
LoRA
Decompression at 1.0
Output
960×544, 121 frames at 24fps
Sampling
8 steps, euler_ancestral
Seed
42
Generated in
7 min
Peak VRAM
11.1 GiB
Ran on
RTX 3060, 12GB (local)

What was wired in

  • Blocky market
    Guide video
    Blocky marketA generated market clip, re-encoded at 120 kbpsDownload MP473 KB
Works

I re-encoded a generated market clip at 120 kbps to break it into blocks. The LoRA removes the blocks, and the fruit and the crowd come back clean and close to the original.

Reference shows a market stall under a green and white striped awning where a vendor with a grey beard in a blue apron is arranging oranges, red apples and yellow bananas in wooden crates while a customer in a dark red headscarf reaches for lemons, with a busy market crowd behind, heavily compressed with visible macroblocking, chroma bleed, and ringing artifacts. Edited shows the same scene restored to high quality with sharp detail, clean edges, and no compression artifacts. ENHANCE QUALITY a market stall under a green and white striped awning where a vendor with a grey beard in a blue apron is arranging oranges, red apples and yellow bananas in wooden crates while a customer in a dark red headscarf reaches for lemons, with a busy market crowd behind. Subject identity, framing, and background geometry are identical to the reference; only compression artifacts and image quality differ between reference and edited.

Notes

It came out so close to the original that I tried footage that was never clean next. The trigger is ENHANCE QUALITY, and it only helps footage that is badly damaged.

The Decompression LoRA on Hugging Face

Run it yourself

Download everything for this test (.zip, 83 KB). It holds the workflow, every input it loads and a README. Copy the inputs into ComfyUI/input, put the models below in place and drag the workflow onto ComfyUI.

The Lightricks repositories (the LTX 2.5 base model and the LoRAs) are gated. Log in to Hugging Face and accept the terms on each repository page before downloading.

Custom nodes: ComfyUI-LTXVideo at commit 3bf3ca6. Everything else is part of ComfyUI 0.37.0 (commit 0f74f7fb, frontend 1.53.6).