A workflow is scenes as columns of nodes
The workflow canvas lays scenes as columns of Image, Video and Voiceover nodes. Edit one prompt, rerun just that node, then save the recipe.
A workflow on the canvas is scenes laid out as columns. Each scene is a stack of nodes — typically Image → Video → Voiceover. Open the workflow canvas. Describe the reel in one sentence. The agent drafts the columns. You edit a prompt on one node and rerun just that node. You do not regenerate the whole show because the coffee steam was wrong.
This is not the video editor. The editor is a timeline and a chat. The canvas is a recipe: columns you can save, rerun, and put on a schedule. Same account as the Android app on your phone. The long build happens here.
Describe it in one sentence
Sign in at app.versely.studio on the account you already use on Android, or create the free one. Open Workflows. Talk to the agent on that screen. One sentence is enough.
A single plate sounds like: "a neon city image, then animate it into a 9:16 video." That drafts one column — Image, then Video. For a reel you actually post, name the scenes, the frame, and the voice:
"A three-scene 9:16 morning routine reel. Scene 1 is a bedside alarm. Scene 2 is a coffee pour. Scene 3 is leaving the apartment. Voiceover on each scene."
The agent lays out the columns. Read them before you run anything. You want three columns, not one long prompt pretending to be three shots. Each column should show an Image node, a Video node hanging off that still, and a Voiceover node.
Do not open create a multi-scene AI movie and expect a canvas you can rerun next Tuesday. That job renders a film. This screen stores a recipe. Pick the canvas when you will make the same shape again.
Three columns for a morning routine
Treat the draft as a storyboard you can edit, not as a finished cut. Click into each node's prompt and write the shot you would actually accept.
| Column | Image node | Video node | Voiceover |
|---|---|---|---|
| 1 — alarm | Phone-lit bedside, cheap clock at 6:12, rumpled duvet, 9:16 | A hand hits snooze, slight shake | "I used to lose the first hour of every day." |
| 2 — coffee | Small kitchen, kettle just off the boil, ceramic mug, morning window, 9:16 | Pour into the mug, steam, no faces | "Kettle, mug, five minutes. That is the whole system." |
| 3 — leave | Apartment door, keys in the bowl, bag on the hook, 9:16 | Pick up the keys, open the door | "Do the first thing before the phone." |
Keep the person out of the stills if you do not have a reusable face. Hands, steam, a door survive a weekly rerun. A generated "same woman" will not.
Run the nodes in the order they depend on each other. Scene 1 Image first. Wait until there is a still. Then Video, then Voiceover. Then Scene 2, then Scene 3. Do not fire every Video node the moment the columns appear.
The Video node on a connected still is the same kind of work as turn a photo into a video. The Voiceover node is write and generate a voiceover for that scene's line. Watch each still before you animate it. A 6:12 that rendered landscape, a mug that looks like a cocktail — fix the Image prompt and rerun that node. Animating a wrong still wastes a Video run.
Image before video. Two videos before a merge.
Two rules. Most failed first runs break one of them.
A connected Image node needs an image before its Video node runs. The Video node is image-to-video. It has nothing to drive if the still is empty, still generating, or left over from a node you rewired. Run Image. Confirm the file is on the node. Then run Video. If you change the Image prompt later, rerun Image, then Video. The motion is of that still, not of the sentence above it.
A Video node with no Image attached is text-to-video. That is allowed. It is not this chain. If the pour has to match the mug in the still, attach the Image node and wait.
Merging needs at least two videos. The reel is Scene 1's video plus Scene 2's video plus Scene 3's video, laid end to end. One finished column is a clip. Two finished columns can merge. Three is this example. Do not look for a merge on a canvas that has only produced one file.
Voiceover is per scene. Write the line on the Voiceover node in that column. If Scene 2's line is wrong and Scene 1 is fine, you will be glad they are separate nodes.
When a node hangs, that is a run-history problem, not a reason to duplicate the workflow. Check or resume a stuck workflow run is the named job. Open the history for this run. Resume the node that stopped.
Edit one prompt. Rerun only that node.
This is the point of columns. A movie job regenerates the whole reel because one beat was wrong. The canvas does not.
Scene 2 came back as espresso when you wanted a pour-over. Open the Image node on column 2. Change the prompt to "pour-over, paper filter, gooseneck kettle, ceramic mug." Rerun that Image node, then the Video node under it, because the still changed. Leave Scene 1 and Scene 3 alone.
Scene 1's voiceover is too slow, or it says "hour" when you meant "morning." Open the Voiceover node on column 1. Edit the line. Rerun that node. The alarm still and the snooze motion stay.
The Video on Scene 3 cropped out the keys. Edit the Video prompt: "hands pick the keys out of the bowl, then open the door, 9:16, keep the bowl in frame." Rerun Video on column 3. Do not touch the Image node if the still was right.
What you do not do:
- Rerun the whole workflow because one line of VO was wrong.
- Duplicate the canvas into "morning routine v2" every time a still misses.
- Change three prompts, then rerun only the last node.
Every run has a run history. Point at the Image run that produced the mug you kept and the Voiceover run you just replaced. If you cannot tell those apart, you will rerun the wrong node.
Every plan gets the same models and editor. The canvas is not a Pro screen. Running a node spends credits the way any generation does. Saving the layout does not. Do not type a credit number into a prompt. Read what the run actually bills. Merge once the three beats are the ones you would post.
Save the recipe. Run now. Put it on a clock.
Saving a workflow stores a reusable recipe. It does not generate episode two. Save and reuse a video workflow is the named job if you would rather say it to the agent. The split is build it once: save_workflow is not the generate button. Save when the three columns, the prompts, and the node graph are the show you want to repeat. Do not save an empty draft.
Run now makes the agent produce the next episode. That is the generate: new stills, new motion, new VO against the same columns. Scene 1 is still an alarm, Scene 2 is still a pour, Scene 3 is still a door. Click Run now when you want another morning routine, not when you wanted to fix one prompt. A prompt fix is rerun-that-node. A new episode is Run now.
A recurring series is the same recipe on a schedule you set through the agent. Run a recurring video series is the named job. Tell it the cadence and which connected accounts should receive the file. The series posts on that schedule. You can pause and resume it.
Playbooks are the other door: twelve launchable recipes that prefill the agent. The headline one is a story-time slideshow series. Use a playbook when you want their recipe. Use this canvas when the recipe is yours.
Once a run has produced the reel, schedule it from Auto Post. Rescheduling a slot or editing a caption later does not charge you again. The credits were the generations on the nodes, not the calendar click.
Draft three columns, then save them
Three actions. Do them on a reel you would actually post, not on a neon city you will never schedule.
- Open the workflow canvas. Type the morning-routine sentence: three scenes, 9:16, alarm then coffee then the door, voiceover on each. Let the agent draft the columns. Fix the prompts in the table above before you run a node.
- Run Scene 1 Image, wait for the still, then Video, then Voiceover. Repeat for Scene 2 and Scene 3. Merge only when you have two or more videos. If a node sticks, resume that run from history. If one still is wrong, rerun that node, not the workflow.
- Save the workflow as a recipe. Click Run now when you want episode two. If you want a cadence, ask the agent to run it as a recurring series on the accounts you already connected.
The canvas is the show. Run now is an episode. A node rerun is a correction. Keep those three jobs separate.