✓
Runs with no disk access
Marginal recommendation (largest combo with no disk access): 864×480 · 8s, estimated 4 min 14 s
RTX 4070 12GB32GB RAMPruned INT8 ConvRot (recommended)LightX2V Turbo 8-step v1.0SageAttentionCalibrated · 4 anchors
VRAM peak
11.2 GB
RAM needed / available
21 / 29 GB
SSD read per step
12.0 GB
Pagefile write / clip
0
SSD wear / 100 clips
0%
⇩ Download current main model⇩ Model library (quant / acceleration / fine-tunes)Check the precision tier matches your VRAM/RAM before downloading
16 combos will use virtual memory|1 combos exceed limits and cannot run
Runtime config calculator
Click a cell for time breakdown, resource usage and SSD wear details
Aspect ratio
Per-clip time budget
Recommended = highest-quality "no-disk" combo within budget (threshold point)
← Scroll the table sideways for other durations →
| Resolution \ Duration | 5 s | 8 s | 10 s | 15 s |
|---|---|---|---|---|
416 × 256 0.1MP · flicker-prone | ||||
864 × 480 0.4MP | ||||
960 × 544 0.5MP | ||||
1152 × 640 0.7MP | ||||
1376 × 768 1.0MP · above native | ||||
1920 × 1088 2.0MP · above native |
No disk spill (safe zone) Uses pagefile / heavy SSD reads (slower) Severely over limits (pagefile too small / read bottleneck / over 90 min)Peak write 0.0 GB/clip · click a cell for details
Orange tiers mark above-native canvas (H3 native cap: 768 short side / 1344 long side, i.e. 0.98MP). Higher pixel counts (1.0-2.0MP) require the official H3-Regenerate-2K (closed-source, API only) or a community "latent upscale + second pass" workflow; estimates exclude the extra time and VRAM of a second pass. Sizes follow the official ComfyUI ResolutionSelector algorithm (multiple=32).
864×480 · 8s193 frames · 20,565 tokens · 15.68s/step
RunnableTime breakdown (total 4 min 14 s)
Model load12 s
Text encoding39 s
Sampling (8 steps)2 min 5 s
VAE decode1 min 11 s
Fixed overhead6 s
Resources & wear
Peak VRAM
11.2 GB
RAM needed / available
21.0 / 29 G
Weight offload
14.8 G / step
SSD read per step
12.0 G
Pagefile write / clip
0 (no disk spill)
SSD wear / 100 clips
0%
- · Not enough RAM to cache all weights: ~35GB must be read back from SSD (~12.0GB/step, read-only)
- · Partial weight offload: ~14.8GB moved over PCIe per step
Self-calibration (optional)
Run once at 832×480 / 5s with current settings, enter your measured s/it; the whole table scales proportionally. Current model prediction: 11.58 s/it