# Flow Studio > A visual workspace connecting references, prompts, versions and feedback around Google Flow. Personal tooling · Creative automation Category: Tools & research [Open in the interactive portfolio]() [Français]() ## At a glance | Metric | Value | | --- | --- | | My role | Application & automation | | Workspace | Graph · versions · review | | Generation | Google Flow | | Example | Existing Coco Romy production | ## Keep the creative thread Generating images and videos involves more than a prompt. References, earlier results and requests for changes need to stay connected. I built Flow Studio to keep that work in one visual workspace, with a browser adapter for Google Flow. These captures use Coco Romy, an existing generated production in my workspace. They show a real project with its images, prompts, versions and clips, rather than a reconstructed interface. ### Images and media [The graph keeps the relationships visible; the inspector provides the selected node’s complete content.]() Flow Studio graph connecting images, prompts and video shots, with the selected image and its instructions in the inspector. Credit: Flow Studio · existing Coco Romy production. Captured 9 November 2026. ## Connect instructions to their results The work is organized into nodes for prompts, references, images, videos and review steps. An image keeps its instructions and references; the connected video shot retains the context of the image it uses. The aim is to revisit a decision without having to reconstruct which file belonged to which version. The graph makes the sequence readable, while the store retains identifiers, settings and downloaded files. TypeScript · React · React Flow / XYFlow · File-backed state · Google Flow From brief to generated shot 1. **Prompt** — Define the scene, references and what must be preserved. 2. **Image** — Review the result and retain its versions. 3. **Video** — Connect the shot to its image and animation instructions. A simplified reading of the workflow visible in the existing project. ## Revise an image without losing earlier versions In this example, the requested change was specific: swap the characters’ positions on the scooter while preserving their identities, clothes and lighting. The approved version places Coco at the handlebars and Romy behind him. History retains the first version, the agent’s revision and the approval state. Earlier instructions remain available to inspect, compare as text and restore without erasing what happened afterwards. ### Images and media [Version 1, preserved in history.]() Earlier generated version with Romy at the front of the scooter and Coco behind. Credit: Coco Romy · Google Flow-generated image, existing project archive. [Approved version 2: the seating changes while the scene remains recognizable.]() Revised version with Coco driving and Romy behind him in the same dawn street. Credit: Coco Romy · Google Flow-generated image, existing project archive. [The earlier result and its instructions remain accessible from the same node.]() History tab showing both image versions and the earlier instructions. Credit: Flow Studio · existing Coco Romy production. Captured 9 November 2026. ## Turn feedback into actionable work Review offers three decisions: approve, comment or request changes. Submitted feedback is tied to the version being reviewed and becomes a task the agent can read. A revised result returns as a new version of the same node. That lets me specify what should change and what should remain consistent. Review history and task acknowledgements keep the feedback connected to the revision. ### Images and media [A demonstration draft entered for the capture and then cleared. No new feedback was submitted to the existing production.]() Review tab with a Portfolio demo draft and Approve, Comment and Request changes actions. Credit: Flow Studio · existing Coco Romy production. Captured 9 November 2026. ## Carry the image into a video shot The video shot keeps its image and motion instructions close at hand. Its detail panel makes it possible to inspect the scene and continuity requirements before approving the result. The clip below comes from an existing Coco Romy generation. It runs for about eight seconds, in portrait 720 × 1280 at 24 frames per second. This take is still awaiting review in the workspace; it is shown as a concrete output example rather than an approved final film. ### Images and media [The clip, its instructions and its Ready for review state in one view.]() Selected video node with playback paused at one second and its animation instructions in the detail panel. Credit: Flow Studio · existing Coco Romy production. Captured 9 November 2026. [Existing Coco Romy scene 13, second take. Click to play.]() Generated clip of Coco and Romy on a scooter in a Riviera street at dawn. Credit: Generated with Google Flow, 3 November 2026. Optimized portfolio copy; scene and audio retained. ## Recover a result without regenerating it Generation and downloading are separate steps. If retrieval fails after generation finishes, the system retains the information needed to recover the existing asset before submitting another paid request. This is less visible in the graph, but it matters in a tool that needs to resume after interruptions. Operation states, identifiers and version relationships all serve the same purpose: preserving what actually happened. ## My contribution and the project’s scope My work is the production workspace, persistence, review and automation around Google Flow. The generation models and Google Flow itself are existing tools. Some generation combinations remain partial or depend on the upstream interface and account state. These captures were taken in the live app on 9 November 2026 using an existing production. They demonstrate the workflow without claiming a measured time saving or universal reliability.