First frameJPG · included
Node-based AI workflows · early access
Describe the result. The agent builds the production.
From one short video to fifty campaign variations. Verstak turns the brief into a working graph, estimates it and lets you control every step.
Agent-built graph · typed connections · estimate before run
Motiontext · required
Seedance 2.0 Fast9:16 · 5 sec
Video outputReady · 5 sec
Estimateestimate before run
Brief → working graph
The agent creates and validates the first version.
Run only what changed
A node, branch, group or declared output.
Batch without chaos
Separate status and retry for every item.
Workflow becomes a tool
Publish a safe form without exposing the canvas.
The agent is the starting point
You describe the outcome. It designs the graph.
No need to learn the node catalog before getting value. The agent uses the same validator, snapshots and runtime as the visual editor.
- 01Understands the production brief
Inputs, output format, references and constraints become an explicit recipe.
- 02Builds only valid connections
Types, required fields and model capabilities are checked before anything is saved.
- 03Shows the plan before spending
The run starts only after preflight, estimate and your approval.
You
memory.jpg · 9:16
Turn this portrait into a 5-second 9:16 video. Soft camera push-in, no face changes. Show the estimate first.
memory.jpg · 9:16Workflow assembled
I connected the image and motion brief to Seedance, declared the video output and saved an immutable agent snapshot.
↗Portraitrequired input
¶Motion briefrequired input
▶Seedanceconfigured
□Video outputconfigured
✓Readydeclared output
Snapshot · AGENT · v1
Preflight ready
The product, not a black box
See where every frame came from.
Open the graph when precision matters: inspect inputs, swap a model, select an older result or run only one branch.
↗
Source visualJPG · included
◫
Square1080 × 1080 · cover
◫
Portrait · 1080 × 13501080 × 1350 · cover
◫
Story1080 × 1920 · cover
⌘
File 1:1ready output
⌘
File 4:5ready output
⌘
File 9:16ready output
Product interface demo · responsive canvas
A full production system
Everything after “generate” matters too.
A serious workflow is not a pretty graph. It knows what to run, remembers every exact version and turns repeatable production into a reusable tool.
01
Run at the right scope
Test one node, build everything up to an output or run a whole group. Unrelated ready branches stay untouched.
NodeTo outputGroup1–10 runs
NodeTo outputGroup
Execution plan3 steps · 240 tokens2 ready nodes reused
02
Batch is a first-class workflow
Array or CSV becomes a bounded queue with per-item progress. Retry only failed items and keep every successful result.
Up to 12 items4 in parallelRetry failed
Campaign briefs.csv12 items4 parallel
0102030405060708
03
Every run has an exact version
Immutable snapshots preserve the graph, inputs and selected output. Restore creates a new draft instead of rewriting history.
ImmutableReplayProvenance
v8FULL RUNtoday · 12:41
v7AGENTtoday · 12:32
v6SHAREyesterday · 18:09
04
Save useful pieces once
Keep a model preset, node or connected group on your personal or workspace shelf. Drop it into a new graph with fresh safe keys.
NodeGroupWorkspace shelf
Saved elementsWorkspace
◫Brand imageNano Banana Pro · v3⋮
⌘Reel ending4 nodes · v2⋮
Drag into any workflow
05
Publish a workflow as a tool
Expose only safe inputs and one declared output. Teammates run a simple form while the immutable production logic stays hidden.
Safe formPinned snapshotNo canvas
Photo → motionPublished · v3
2 public fields · 4 internal settings hidden
06
Share without losing control
Owner, editor and viewer permissions for the workspace, plus a pinned read-only link. A legal viewer can duplicate a clean copy.
RolesRead-only linkDuplicate
Five workflows · product templates
Start with a template. Make it yours.
The graph changes with the task; the production logic stays transparent.
3 formats
01 / 05
One visual — 3 formats
Prepare square, 4:5 and 9:16 outputs from one source without calling a model.
01Source02Square03Portrait 4:504Story 9:16053 outputs
Photo to video
02 / 05
Photo to short video
Turn a photo into a five-second vertical clip with a controlled motion brief.
01Photo02Motion brief03Seedance04Video
Style transfer
03 / 05
Transfer a visual style
Apply the visual language of one reference to another image through a safe prompt contract.
01Source02Style reference03Prompt contract04Nano Banana05Output
Compare models
04 / 05
Compare two image models
Run one idea through the fast and Pro models, then compare both results side by side.
01Idea02Reference03Fast + Pro04Compare052 outputs
Batch
05 / 05
Batch image variations
Turn a bounded list of text briefs into a reviewable gallery with per-item status and retry.
01Briefs02Iterator03Generate04Gallery
Built for real production
Creative freedom without losing control.
Every run is reproducible, every decision stays attached to the artifact.
Rerun one branch
Change a prompt, model or reference without paying for the finished parts again.
scene_03 · v5Versions that do not disappear
Pipeline, manual and AI-assisted edits live side by side with their origin.
v3 → HUMAN → v4An honest run receipt
See model, tokens, duration and cost before approval and after completion.
600 tokens · 03:47Run receiptRUN-2841
Approved before start
- Images · Nano Banana
- 120 tokens
- Video · Seedance
- 480 tokens
- Voice · ElevenLabs
- not used
- Assembly and storage
- included
Total600 tokens
One canvas · the right model at every node
Models are tools. The workflow is the asset.
Switch a model in one node without rebuilding the production from scratch.
Nano Banana
Image and references
↗Seedance
Character motion
↗Kling
Controlled video
↗Veo
Video with native audio
↗ElevenLabs
Voice and dubbing
↗Questions
01What is a node-based AI workflow?＋
It is not a separate neural-network model. It is a visual way to connect inputs, AI models, transformations and outputs into one reusable graph that can be inspected, versioned and rerun.
02What is a node in an AI workflow?＋
A node is one explicit operation with typed inputs and outputs: for example a prompt, reference image, image model, video model, crop, iterator or final output. Compatible nodes connect into a production chain.
03How can I connect several AI models in one workflow?＋
Place the models and media operations on the canvas, then connect compatible outputs and inputs. For example: brief → Nano Banana image → Seedance video → declared output. The agent can assemble and validate the first graph for you.
04What are Weavy AI and Figma Weave?＋
Weavy AI was a node-based creative platform that combined generative models and professional editing tools on one canvas. After its acquisition by Figma, the product became Figma Weave.
05Is there a Russian-language alternative to Weavy AI?＋
Verstak can be considered an alternative for controlled AI content production, but it is not a copy of Weavy or Figma Weave. It offers a Russian-language agent, node-based image and video workflows, branch runs, batches, versions and workflow publication as a tool.
06How is Verstak different from Weavy AI or Figma Weave?＋
Figma Weave is centered on an open creative canvas and professional editing. Verstak is agent-first: the agent designs the graph, while the product adds estimates, exact run history, bounded batches and reusable published tools for repeatable content production.
07How is an AI workflow different from a regular AI generator?＋
A generator produces one result from one request. A workflow preserves the entire production recipe, passes artifacts between steps and lets you rerun only the part that changed.
08What can I create with Visual Workflows?＋
You can build image, short-video, voice and assembly pipelines: create visuals from an idea, edit images, compare models, turn a photo into video, continue a scene, assemble clips or generate a bounded batch.
09Which AI models can I use in a workflow?＋
The current product surface includes Nano Banana for images, Seedance, Kling and Veo for video, and ElevenLabs for voice. Model availability and estimate are shown before a run.
10Can I create a video from a photo?＋
Yes. Use the built-in photo-to-video template or ask the agent to assemble it. Add the first frame, describe the motion, choose a video model and declare the final video output.
11Does Verstak support batch AI generation?＋
Yes. An array or CSV can run as a bounded batch with per-item status, controlled parallelism, a result gallery and retry for failed items only.
12Are there ready-made AI workflow templates?＋
Yes. The starter library includes image creation and editing, three-format adaptation, style transfer, model comparison, photo-to-video, scene continuation, clip assembly and batch variations.
13Can a team collaborate on one workflow?＋
Yes. Workspace access supports owner, editor and viewer roles. You can also publish a pinned read-only version by link or let an authorized user duplicate a clean copy.
14Do I have to build the graph myself?＋
No. Describe the result in chat and the agent assembles a workflow. Open the canvas only when you want to inspect or change the production logic.
15Can I change only one scene?＋
Yes. Select a node, create a new version and rerun only its dependent branch. Approved artifacts remain untouched.
16Are these real examples?＋
Yes. The page uses real image and video outputs from the same model stack available in Verstak. The canvas is an interface demonstration of the product workflow.
17Will I see the price before generation?＋
Yes. Verstak shows the model plan and estimate before the run, then records actual cost for every stage.
Idea → workflow → result
Turn your next idea into a workflow.
Describe the result. Verstak will assemble the first production graph and show the estimate.
Get early accessEarly access is opened in order through the Telegram bot.