Choose between a scored single-clip Video Agent, a planned multi-shot Long Video workflow, or a long-video edit; inspect footage, preserve continuity, repair seams, and finish in an editable video composition.
Where: Flowgen → left rail → Add → Video → Agentic Video Gen / Agentic Long Video
Key ideas
- Video Agent optimizes one clip: Agentic Video generates and scores a single clip within the chosen model’s inputs and limits, then can retry against the requested quality target.
- Long Video separates planning from shots: Long Video Gen turns the brief into a shot plan, generates the shots and boundary continuity, checks each shot, and assembles the sequence. Long Video Edit restructures supplied footage rather than starting from an empty brief.
- Continuity tools solve specific problems: Video Analysis describes and transcribes footage; Frame Extract saves exact stills; Video Seams targets a bad cut window. These are distinct tools, not automatic replacements for one another.
- Composition remains editable: Send successful clips into the native Video Editor/composition when timing, layers, transitions, captions, motion, or audio need hand-authored refinement.
Steps
- Choose Agentic Video for one generated clip, Agentic Long Video for a multi-shot sequence, or Agentic Long Video Edit when the flow should restructure existing footage.
- Connect a written brief through a Text source or use the linked project Brand’s relevant fields. Add visual references, source clips, voice, or audio on their specific typed inputs.
- Pick a model that supports the requested duration, aspect ratio, references, audio, and frame controls. Review the quote/Flowies budget and retry policy before running.
- For Long Video, review the shot plan and each shot’s purpose, duration, camera/action, continuity references, and transition before generation.
- Run the agent and inspect per-shot status, score/retry notes, output clip, audio, and boundary frames. Fix a weak shot locally where possible rather than regenerating the entire sequence.
- Use Video Analysis when you need synchronized visual/audio description, transcription, subject position, or dead-part information from an existing clip.
- Use Frame Extract to create exact still references or clean boundaries. Use Video Seams only for a specific discontinuity; connect the clip and the seam/neighbor context the node requests.
- Send the final clip or ordered shot outputs to the native Video Editor/composition. Confirm shot order, clip trims, sound, transitions, captions, frame rate, and total duration.
- Connect the editable composition to Video Render or the relevant output node. Inspect the render target, resolution, frame rate, audio choice, and price before delivery.
Which node for which job?
- Agentic Video: one generated clip with evaluation/retry behavior.
- Agentic Long Video: plan and generate a coordinated shot sequence.
- Agentic Long Video Edit: reorganize or adapt connected source footage into a longer edit.
- Video Analysis: understand a clip; it is not a video generator.
- Frame Extract: produce still frames from a clip.
- Video Seams: repair a localized transition; do not use it as a generic clip restyler.
- Native Video Editor: make a direct, editable timeline/compositing pass.
Control spending and quality
- A retry can make another paid request; use the node’s maximum attempts and target score deliberately.
- Request only the references the model needs and give each a role.
- Avoid paying for a full long-sequence rerun when one shot can be corrected independently.
- Creation/planning is distinct from execution; run paid generation only with user approval.
Continuity review checklist
- Check character/product identity and wardrobe across adjacent shots.
- Compare action direction, screen position, lighting, lens/scale, and frame boundaries.
- Listen for dialogue/audio continuity and validate captions after timing changes.
- Review joins in motion rather than only checking still thumbnails.
Tips
- Keep an approved shot plan and named references connected so a targeted retry has the same context as the original shot.
- After trimming or reordering, rerun boundary analysis and review each seam at normal playback speed.
Limitations and important notes
- Available reference roles, duration, aspect ratios, sound, and retry controls vary by model and node.
- Generated clips may still differ in continuity; QA scores assist review but do not certify a finished edit.
- Frame extraction and analysis output do not automatically repair or alter the source clip.
Troubleshooting
A requested reference role is missing.
Check the selected model and target node; use only the roles supported by that model and connect references to the matching typed input.
A long sequence feels discontinuous.
Inspect shot boundaries, reuse explicit identity/style references, rerun only the affected shot, and use Video Seams for a localized cut repair.
The edit is not changing after Frame Extract or Analysis.
Those nodes produce information/stills. Connect their outputs into the next generation or editing node that should consume them.
Related guides
- Flowgen inputs: connect prompt nodes and the right Brand fields
- Flowgen Video Editor node: native composition inputs, editor surface, outputs, persistence, and MCP
- Video Analysis node: scenes, transcript, subjects, and dead parts
- Extract Frame node: capture a precise still from video
- Repair transitions with the Video Seams node