griptape-ai / griptape-ai/griptape-nodes-library-diffusers

LTX-2.5 video output turns fuzzy and temporally inconsistent after changing a pipeline setting between runs

Open
#64 0 comments 0 reactions 1 assignee View on GitHub

@FN-DiogoTeixeira is already working on this.

Since Sep 17, 2026.

Dominant language
Python
Stars
0
Forks
0
Avg merge
1d 15h
Merged PRs (30d)
11

Description

Problem

Video output from LTX-2.5 comes out fuzzy, soft, and temporally inconsistent (frames disagree with each other) after a pipeline setting is changed between runs. Expected: the same settings produce a clean, consistent result regardless of what was run before in the same session.

Origin

Reported by Diogo Teixeira. Seen on the LTX-2.5 provider in the Modular Diffusion Pipeline Builder, with both the Distilled and Full (SFT) model variants. Reported as reproducible.

Impact

Output is unusable when it happens, and there's no warning — the build reports success. Because it follows a settings change rather than a hard failure, it's easy to mistake for a bad seed or prompt and burn generations chasing it.

Repro / evidence

  1. Build an LTX-2.5 pipeline in the Modular Diffusion Pipeline Builder (either the Distilled or the Full/SFT model variant).
  2. Set memory optimization to Manual.
  3. Run a generation. Output looks correct.
  4. Change one setting only — e.g. the CPU offload strategy (None -> Model, or Model -> Sequential).
  5. Run again with everything else identical, same prompt and seed.

Observed: the second output is fuzzy and frame-to-frame inconsistent.

Also reported to occur when the memory optimization mode is Automatic, and when switching between the Distilled and Full (SFT) model variants.

Screenshot to be attached by the reporter.

Notes

Root cause not established — this is a symptom report. Worth checking as a first step whether the degradation depends on the change between runs at all, or whether starting a fresh engine with the offload strategy already set produces the same fuzziness.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.