Production setup Same model. Same image frequency.
Studio projectConnect the shared helper Refresh saved setup Loading saved settings. No paid jobs or project changes.
Settings and estimate coverage
Length, chapter count and image cadence remain your planning targets. Editing a planner control or dragging a task switches to a manual scenario. Refreshing the saved setup resets manual moves, with Undo available. Generation strategy is shown for comparison; this proposed schedule does not deploy it.
Final video length minutes
Video resolution720p · selected default 1080p Video frame rate24 fps 30 fps · selected default 60 fps
Local video rendererBest tested path for this format NVIDIA GPU · measured 720p30 CPU · compatibility baseline Forecast only. Studio Video output controls the actual helper. Native GPU rendering shares the laptop GPU with voice; cloud image generation remains independent. Loading matched output-format render timing…
Chapters Images / video min
Visual frequency in the finished video. GPU speed updates separately below.
Images / min · continuousHighest tested continuous warm throughput. Excludes startup, Luna waiting and rental idle time.
Use fastest tested settings
Cloud GPU Image timing profileEarlier 1344 × 768 · serial Measured 720p Measured 1080p Conditioning timingFresh prompt + reference encoding Reused conditioning · control Image executionIndependent workers One worker · pipelined requests Multiple workers · buffered requests Worker count / request slotsBest tested count 1 2 3 4 6 Only completed two-round results can calibrate the schedule.
Scheduling modeProposed streaming overlap Installed cloud overlap Fully sequential baseline
Image visual checksOff · technical checks remain Sample images Practical · every image Strict · every image
Sample percentage
Intro seconds Intro images
Capacity & uncertainty
Measured Luna workflowLean Luna · Standard · experimental Classic Luna · Standard · measured Small visual groups · research · fastest observed One storyboard per chapter · research · lower API cost Earlier classic calibration Luna processingStandard · measured timing Flex · half token price · unmeasured delay Fast · twice token price · unmeasured timing Flex time relative to StandardAssumption only. 1 = same time as Standard; Flex may take longer or be unavailable. No measured Flex ETA yet.
Shared Luna API slots1 2 3 4 5 6 7 8 Measured workflows include their tested concurrency once.
Images per scheduling groupDisplay grouping only; inference stays unchanged. Long timelines automatically use larger groups.
Extra image attempts % An allowance, not a measured error rate.
Timing sensitivity ±% Assumed range. Not statistical confidence.
Chapter plans ready before GPU bootHow many ordered plans to wait for, between 1 and your chapter count. Adjusts automatically when chapters decrease.
Earliest GPU boot · project seconds0 = no clock delay. The timed startup option sets this; recalculated suggestions use your current settings.
Prices & cost assumptions · USD GPU price basisPublished Pod listings · checked Oct 9 Historical benchmark rental receipts My selected GPU compute quote Running container disk · GB Selected GPU compute quote / hour · excludes disk Checking rate source…
Official prices · Actual deployment quotes depend on region and offer. No current listing means an explicitly labeled historical fallback. Quote mode applies your entered compute rate only to that GPU; other GPUs use published prices where available.
Director cost sourceAutomatic · measured receipts × selected tier My dollar estimate · already includes tier Director API / 2 story hours USD · selected tier Vision QC / reviewed image USD · historical estimate Shared retained storage / day USD Shared storage days allocated to this estimate Use 20 GB model-volume allowance 20 GB standard network storage is $1.40/month, about $0.0467/day over a 30-day planning month. This button changes the forecast, not your account volume. Add your archive allowance separately to this daily amount. Shared storage continues billing when GPUs are stopped; do not count its full daily charge again for every video made that day.
Run cost and allocated storage are shown separately. Local narration/render electricity, Google archive requests/egress, other retained pods, taxes, currency conversion and unplanned repairs are not included. API cost varies with tokens; free-token eligibility and trial credits are not assumed. This is a forecast, not an all-inclusive invoice.
Reset setup and moves
Schedule Every process GPU benchmarks Recorded runs Evidence & limits
My proposed schedule Dependency-aware simulation.
Undo Repack earliest Play simulation
Focus chapterAll chapters ZoomFit 2× 4× 8× 24× 60× · seconds
Drag a task left or right. Dependencies and shared resources stay enforced. ← → move 1s · Shift 30s
Select a task to inspect its exact inputs, resources, timing and blockers.
Where time goes Work ≠ elapsed time
Overlapping right now Simulation not running Play the schedule to watch independent tasks run together.
Can these two run together? First task Second task
Earlier serial reference profiles Four matched trials, two historical profiles. Choose a GPU to compare the complete schedule.
Complete schedule comparison Same chapter count, checks, intro, model and scheduling rules. Different historical hardware timings.
More VRAM can fit more models. It does not establish faster simultaneous inference. Measured worker groups preserve the tested model-copy and request-slot counts. Untested counts have no timing estimate.
Loading the director study…
Evidence before promises. Measured totals, component assumptions and unsupported overlaps stay separate.
QuestionTemplate