Workflow

AI Video Workflow

You are not making one clip — you are running a pipeline. Reference, prompt, model, take, follow-up shot, audio, lip-sync, export. Astorie gives you a node-based canvas where every step in your AI video workflow lives next to the last one, models chain into each other, and the output drops into Premiere or DaVinci without leaving the page.

What this feature solves

Real AI video work is not one prompt and one download. It is a sequence of decisions — which model for the establishing shot, which for the close-up, where to lock the character, when to switch from image to video, how to feed the audio in, and how to land the whole thing in an editor. Doing that in browser tabs is brutal: you copy outputs between tools, lose the lineage of which prompt produced which clip, and rebuild the same setup every time you start a new spot.

The deeper problem is reuse. Once you crack a workflow that produces a campaign-quality cut — say, image-to-video on Seedance 2, chained into a follow-up shot on Kling 3, with ElevenLabs voice and lip-sync — you cannot save it. Tab-based tools force you to recreate the chain manually for every new project, which kills the compounding value of a working pipeline. Production teams need templates, not re-runs.

Then there is the export problem. Even when the generations are clean, getting them into Premiere Pro or DaVinci Resolve as a real timeline (not a folder of MP4s with mismatched frame rates) is its own multi-hour job. Every transcode is a chance to lose quality, every codec mismatch is a re-import. A workflow that ends at MP4 is half-done.

Why Astorie is different

Astorie is a canvas, not a tool stack. Every node — image, video, audio, script, text, tool — sits on the same surface, and connections between nodes are real wiring. The output of an image node is the literal input of a video node. The output of a video node feeds a lip-sync node, which feeds a sequence builder, which exports as one timeline. You see the entire pipeline at a glance, and you edit it like a graph, not like a stack of browser sessions.

Model chaining is the unlock. You are not stuck in one provider — Sora 2 for the wide, Seedance 2 for the close-up, Kling 3 for the camera move, Veo for the photoreal cutaway. Each node picks its own model, and the canvas treats them as one pipeline. Fanout (run the same input through three models) and chaining (run the output of one model into another) compose freely. That is impossible inside a single-model SaaS tool.

Workflows are saveable artifacts. Build a pipeline once — image-to-video on Seedance, chain to a follow-up on Kling, layer ElevenLabs voice, lip-sync, export — and save the whole canvas as a template. Next campaign, swap the inputs and re-run. The compounding value of a node-based workflow is that the second project is faster than the first, and the tenth is faster than the second. Tab-based tools can never give you that.

Common use cases

Storyboard-to-clip pipeline for ad concepts

Generate frames in image nodes, fan them into video nodes for animation, and assemble the cut on one canvas before client review.

End-to-end product campaign builder

Chain hero shot, b-roll, cutaways, and voiceover into one reusable template you can rerun for every SKU launch.

Multi-shot scenes with character continuity

Pin a reference character across image and video nodes so the subject stays identical from shot one to shot eight.

Short film pre-viz to picture lock

Move from script node to storyboard frames to animated shots to NLE export — all in one canvas with full lineage tracking.

Agency template library for repeatable deliverables

Save winning pipelines as canvas templates and let the team rerun them for every new client brief without rebuilding from scratch.

A/B testing creative variants for paid social

Fan one storyboard into multiple model and prompt combinations, then export the strongest takes for ad-platform testing.

Recommended model stack

How the workflow works in Astorie

  1. 1

    1. Open a new canvas and add your inputs

    Drop reference images, scripts, or briefs onto the canvas as image and text nodes. These are the source of truth for everything downstream — keep them organized so the pipeline reads left to right.

  2. 2

    2. Add the first generation node

    Connect inputs into an image or video node. Pick the model that matches the shot type — Seedance 2 for product fidelity, Kling 3 for camera moves, Sora 2 for long takes. Run and review.

  3. 3

    3. Chain the next stage

    Wire the output of stage one into the next node — a follow-up video shot, a lip-sync node, an audio node, an upscaler. The canvas remembers the lineage so swapping the source re-renders everything downstream.

  4. 4

    4. Fan out for variants

    Duplicate generation nodes and swap models or prompts to run multiple takes from the same input in parallel. Compare results side by side without leaving the canvas.

  5. 5

    5. Assemble into a sequence

    Connect the chosen takes into a sequence builder. Order shots, set transitions, and preview the full cut before export.

  6. 6

    6. Export to your NLE or save as a template

    Push the sequence into Premiere Pro, DaVinci Resolve, or Final Cut Pro with NLE export, or save the entire canvas as a template for the next campaign.

Example workflow

A creative agency is producing a 30-second product launch spot with eight cuts. The producer drops the brand-approved hero photo into an image node and the script into a text node. From the hero image, they wire into three parallel Seedance 2 video nodes — one per hero cut — and use Kling 3 for the moving establishing shot. A reference character is pinned and fed through Nano Banana 2 for talent stills, which then chain into Hailuo for spokesperson motion. ElevenLabs handles voiceover, lip-sync wires audio into the spokesperson cuts, and the entire eight-shot sequence routes into a sequence builder. NLE export drops the timeline into Premiere ready to grade. The whole canvas is saved as the agency template — next SKU, swap the inputs and re-run.

Tips and common mistakes

Tips

  • Build left-to-right. Inputs on the left, generations in the middle, sequence and export on the right. The visual flow makes complex pipelines readable.
  • Save canvases as templates the moment a workflow ships a deliverable. The compounding value is in reuse.
  • Use color or labels on nodes to mark hero shots, b-roll, and cutaways — large pipelines get hard to navigate without it.
  • Chain reference images forward so character and style persist across multiple video nodes — don't re-upload at every stage.
  • Keep one canvas per project. Forking a working canvas for variants beats restarting from scratch.

Common mistakes

  • Treating the canvas as a notepad instead of a pipeline. Discipline the layout and the workflow gets faster every time you reuse it.
  • Using one model for every shot. Different shots need different engines — chain models per shot type, do not pick a favorite.
  • Skipping the sequence builder and exporting clips one at a time. You will rebuild the timeline manually in your NLE — wasted hours.
  • Forgetting to save the canvas as a template. A pipeline that took two days the first time should take twenty minutes the second.
  • Hard-coding prompts into nodes when you could parameterize them. Templates that accept variables scale across teams; one-off nodes do not.

Related how-to guides

Related models and tools

Related features

Related docs

Related reading

Comparisons

Frequently asked questions

How is this different from ComfyUI?

ComfyUI is a powerful node graph for image generation but it is single-machine, single-provider per node, and built around Stable Diffusion checkpoints. Astorie is multi-provider out of the box (Sora, Seedance, Kling, Veo, Runway, etc.), runs in the cloud, and is built around production video workflows including NLE export — not just image generation.

Can I save a workflow and reuse it across projects?

Yes. Save the entire canvas as a template, then duplicate it and swap the input nodes for the next project. The model chains, prompts, and sequence settings persist. This is the core unlock for agencies running repeat client work.

Do I need to learn the node interface from scratch?

If you have used Figma, Final Cut, or any timeline editor, the canvas reads naturally. Nodes are blocks, connections are wires, output flows left to right. Most users build their first end-to-end pipeline in under an hour.

Which models should I chain together for a product spot?

For most product spots, start with Nano Banana 2 or Midjourney for storyboard frames, Seedance 2 or Kling 3 for animation, ElevenLabs for voiceover, and lip-sync for any spokesperson cuts. Save that chain as your template and adjust per brand.

How does this compare to Runway?

Runway is a strong single-provider tool with its own video models. Astorie is the canvas layer above the providers — you can run Runway Gen-4 inside Astorie alongside Sora, Seedance, Kling, and Veo, fanning and chaining across all of them. The workflow lives in Astorie; the models are pluggable.

Will my pipeline output drop into a real NLE timeline?

Yes. NLE export renders sequences at clean frame rates and codecs that Premiere Pro, DaVinci Resolve, and Final Cut Pro open natively. The eight clips in your spot land as one cut, not eight orphan files.

Build it on the canvas

Open Astorie and wire this workflow up in minutes. Free to start — no card required.