Skip to content

Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS

Keep the LTX conditioning round trip off the Windows helper

Claude38c054236877
diff --git a/server/services/videoGen/conditioningMedia.js b/server/services/videoGen/conditioningMedia.js
index c39c3c32a9..48a02ccde5 100644
--- a/server/services/videoGen/conditioningMedia.js
+++ b/server/services/videoGen/conditioningMedia.js
@@ -87,7 +87,9 @@ export async function prepareVideoConditioningMedia({
}
const ffmpeg = (sourceImagePath || lastImageWillBeUsed || hasMultiKeyframes) ? await findFfmpeg() : null;
const ffprobe = model.runtime === 'minimax_h3_ref2va' ? await findFfprobe() : null;
- const conditioningCrf = CONDITIONING_CRF_BY_RUNTIME[model.runtime || 'mlx_video'] ?? null;
+ // Keyed on the runner, not just the runtime name: a model that routes to
+ // generate_win.py also reads as mlx_video, but that pipeline is LTX-Video 0.9.
+ const conditioningCrf = routesToWindowsHelper(model) ? null : (CONDITIONING_CRF_BY_RUNTIME[model.runtime || 'mlx_video'] ?? null);
const resizeImage = async (srcPath, tag) => {
if (!srcPath || !ffmpeg) return { resolved: srcPath, tempPath: null };
const resizedPath = join(tmpdir(), `resized-${tag}-${jobId}.png`);