If you tried 3.0 early and hit "error generating video" — or went back to 2.0 to get image-to-video working — that's fixed now. Update in Pinokio, and if it still errors once, click Install again (it refreshes the bundled Python packages and clears stale installs).
What 3.0 added over 2.x
- Characters — train a face + voice in-app, then generate talking clips
- Audio-to-video — drive a clip from an audio track
- Image Studio — generate stills
- Quality tiers — Quick / Balanced / Standard (fast Q4) and High (Q8 + TeaCache)
- RAM-aware tiers — the panel reads your Mac's memory and keeps big renders out of swap
Fixed across 3.0.1 → 3.0.8
- The generation crashes (the
fps / frame_rate / "No module" errors) — gone. That's why some of you fell back to 2.0; you don't need to anymore.
- The image-to-video "mosaic" / grid output on some Macs — fixed. It was a macOS Metal timing issue, not low RAM.
- Boost/Turbo speed restored, CivitAI browser SSL fixed, and the whole LTX engine re-pinned so an upstream change can't silently break it again.
Common questions
- Near-silent audio? Describe the sound in your prompt ("low whispered chant, ember crackle") — audio is generated from your words, jointly with the video.
- Clip length? Renders are a few seconds at a time by design — use Extend to add more.
- 32 GB Mac? The biggest modes cap at 768px to stay out of swap; run an upscaler after if you want 720p+.
Thanks for the detailed reports — they drove every one of these fixes.
If you tried 3.0 early and hit "error generating video" — or went back to 2.0 to get image-to-video working — that's fixed now. Update in Pinokio, and if it still errors once, click Install again (it refreshes the bundled Python packages and clears stale installs).
What 3.0 added over 2.x
Fixed across 3.0.1 → 3.0.8
fps/frame_rate/ "No module" errors) — gone. That's why some of you fell back to 2.0; you don't need to anymore.Common questions
Thanks for the detailed reports — they drove every one of these fixes.