SCAIL-2 character swap fails on H100: 5.5-hour run yields blurry frames despite success status
ebubu03 · reddit · 2026-09-16
A Reddit user ran SCAIL-2 (Wan2.1 14B-based) character replacement in ComfyUI on a RunPod H100 and hit a baffling failure: a 3-minute test (576×1024, 81 frames, turbo + LightX2V, 6 steps) worked fine, but the full run (768×1344, 237 frames, turbo off, LightX2V bypassed, 40 steps, CFG 5) took 5h28m and $19.18, producing only blurry brown/gray frames — while ComfyUI still reported executionsuccess.
Key details:
- All 237 frames processed in one pass instead of chained 81-frame segments; resolution, frame count and turbo were changed together, so no controlled comparison
- SAM3 tracking finished in 23s, then logs stalled at "31270MB Staged, 400 patches attached" while sampling continued in the UI
- Telemetry showed 100% GPU utilization, 56/80GB VRAM, 698W, 57°C, and no recorded exceptions
- One suspicious wiring detail: BasicScheduler receives the model directly from the LightX2V node, while SamplerCustom gets the turbo/non-turbo model via ModelSamplingSD3
The author asks whether chained segments are required, whether the non-turbo config is sound, whether the issue is sampling instability, conditioning/masks or VAE decoding, and for a quick diagnostic to distinguish bad latents from a VAE problem.
More from Multimodal
- Audio8-TTS-Realistic-Air, a zero-shot voice-cloning TTS model, trends on Hugging Face — Edge0 · 2026-11-02
- Mistral explains why its model-built solver is a first step toward new drugs and batteries — MistralAI · 2026-11-02
- Mistral Large 4 writes a quantum chemistry SCF solver from scratch with one short prompt — MistralAI · 2026-11-02
- Gorgeous dithered pixel art collage serves up a 'great morning' — philipvollet · 2026-11-02
- Clown Prophecy moves to new repo, now uses first clip as reference to fix H3 quality drift — reeight · 2026-11-02
- A 25-year design veteran asks: who still mixes pro software with AI tools? — Opening-Knee-5913 · 2026-11-02