What are you making?
Start with a media task. Find saved resources, check their limits and let your agent apply them.
Connect your agent · Add Veyrul to a workflow · Collaborate
Keep word captions aligned after cuts and speed changes
Reuse timing code and edge cases for clipped words, repeated source intervals and rational playback speeds.
Normalize spoken audio without guessing the delivery target
Start from a loudness target, measure the input and verify the encoded result. Keep noise cleanup and loudness adjustment separate.
Select a usable audio stream from a mixed Apple MOV
Inspect stream metadata before choosing between an ordinary audio track and an APAC stream your decoder may not support.
Turn beat times into a bounded video cut plan
Use saved timing helpers to propose cuts while keeping musical judgment, frame snapping and source limits explicit.
Compose FFmpeg filters without losing an earlier operation
Reuse a small filter-composition example and its checks when an agent needs several video operations in one command.
Choose a cutout mode for hair, fabric and product edges
Compare alpha-only assembly, edge color correction and matting before promising that original RGB values will be preserved.
Plan masked video editing with a VACE workflow
Find the documented input, mask and frame constraints before attempting object removal or another regional video edit.
Preserve a character across camera angles
Use the proposed reference-to-keyframe-to-animation route with explicit checks for identity, costume and unseen details.
Check whether a video workflow fits your GPU
Separate model-card hardware claims from offloaded template behavior before downloading weights or promising a memory fit.
Bring a glTF asset into a React Three Fiber scene
Find gltfjsx usage and distinguish shared geometry, transformed instances and animation support.
Plan accessible motion and choose a Lottie player
Use motion-design guidance alongside player-specific support notes and a reduced-motion alternative.
Review a manuscript without silently accepting every edit
Keep proposed changes, author acceptance and a Word redline as distinct steps with preserved originals.
Prepare AI cover artwork for a print-cover handoff
Separate artwork approval from trim size, spine, bleed, safe area, barcode and export requirements.
Turn a script into a reviewable AI shot plan
Compare documented script, character and shot interfaces before commissioning generation.
Align podcast words and speakers with WhisperX
Check language alignment, missing word timings and diarization access before treating a transcript as ready for captions.
Choose between cropping and outpainting a vertical video
Estimate what a 9:16 delivery removes or invents before selecting a reframing method.
Plan a consistent virtual character across posts and shots
Compare documented reference and frame-relay approaches, and separate creator reports from verified identity consistency.
Find a useful reference for an AI commercial concept
Use saved references to compare visual decisions, production approaches and unanswered questions about execution.
Keep AI-assisted music editable as symbolic MIDI
Preserve note events, timing and track structure while keeping MIDI authoring distinct from rendered audio quality.
Turn an AI illustration into an editable SVG
Compare tracing with direct SVG generation and inspect the actual structure before calling the result editable.
These starting points link to saved evidence. A documented method, a tested fixture and a proven production result are different things. Each task makes its current limits explicit.